X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=a93f4f67629781d711b5afc8c7751d6e941fff62;hp=e4b9af4a3ed8451b18fc2f4da09c0e806ab51a5e;hb=b806a81070940ce13e6257aa99de18a71f08706b;hpb=ccabce23bd9e366c345c852f565766a799f61238 diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index e4b9af4..a93f4f6 100644 --- a/lustre/osd-ldiskfs/osd_handler.c +++ b/lustre/osd-ldiskfs/osd_handler.c @@ -39,6 +39,7 @@ #define DEBUG_SUBSYSTEM S_OSD +#include #include #include #ifdef HAVE_UIDGID_HEADER @@ -76,6 +77,9 @@ #include +/* Maximum EA size is limited by LNET_MTU for remote objects */ +#define OSD_MAX_EA_SIZE 1048364 + int ldiskfs_pdo = 1; module_param(ldiskfs_pdo, int, 0644); MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations"); @@ -136,12 +140,12 @@ int osd_trans_declare_op2rb[] = { static int osd_has_index(const struct osd_object *obj) { - return obj->oo_dt.do_index_ops != NULL; + return obj->oo_dt.do_index_ops != NULL; } static int osd_object_invariant(const struct lu_object *l) { - return osd_invariant(osd_obj(l)); + return osd_invariant(osd_obj(l)); } /* @@ -149,18 +153,19 @@ static int osd_object_invariant(const struct lu_object *l) */ static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o) { - struct osd_thread_info *oti = osd_oti_get(env); - return oti->oti_w_locks > 0 && o->oo_owner == env; + struct osd_thread_info *oti = osd_oti_get(env); + + return oti->oti_w_locks > 0 && o->oo_owner == env; } /* * Concurrency: doesn't access mutable data */ static int osd_root_get(const struct lu_env *env, - struct dt_device *dev, struct lu_fid *f) + struct dt_device *dev, struct lu_fid *f) { - lu_local_obj_fid(f, OSD_FS_ROOT_OID); - return 0; + lu_local_obj_fid(f, OSD_FS_ROOT_OID); + return 0; } /* @@ -178,9 +183,10 @@ static struct osd_idmap_cache * osd_idc_find(const struct lu_env *env, struct osd_device *osd, const struct lu_fid *fid) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_idmap_cache *idc = oti->oti_ins_cache; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_idmap_cache *idc = oti->oti_ins_cache; int i; + for (i = 0; i < oti->oti_ins_cache_used; i++) { if (!lu_fid_eq(&idc[i].oic_fid, fid)) continue; @@ -197,8 +203,8 @@ static struct osd_idmap_cache * osd_idc_add(const struct lu_env *env, struct osd_device *osd, const struct lu_fid *fid) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_idmap_cache *idc; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_idmap_cache *idc; int i; if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) { @@ -247,10 +253,16 @@ osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd, if (idc != NULL) return idc; + CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n", + osd->od_svname, PFID(fid)); + /* new mapping is needed */ idc = osd_idc_add(env, osd, fid); - if (IS_ERR(idc)) + if (IS_ERR(idc)) { + CERROR("%s: FID "DFID" add id map cache failed: %ld\n", + osd->od_svname, PFID(fid), PTR_ERR(idc)); return idc; + } /* initialize it */ rc = osd_remote_fid(env, osd, fid); @@ -282,8 +294,8 @@ static int osd_idc_find_and_init(const struct lu_env *env, struct osd_device *osd, struct osd_object *obj) { - const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu); - struct osd_idmap_cache *idc; + const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu); + struct osd_idmap_cache *idc; idc = osd_idc_find(env, osd, fid); LASSERT(!IS_ERR(idc)); @@ -298,10 +310,16 @@ static int osd_idc_find_and_init(const struct lu_env *env, return 0; } + CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n", + osd->od_svname, PFID(fid)); + /* new mapping is needed */ idc = osd_idc_add(env, osd, fid); - if (IS_ERR(idc)) + if (IS_ERR(idc)) { + CERROR("%s: FID "DFID" add id map cache failed: %ld\n", + osd->od_svname, PFID(fid), PTR_ERR(idc)); return PTR_ERR(idc); + } if (obj->oo_inode != NULL) { idc->oic_lid.oii_ino = obj->oo_inode->i_ino; @@ -319,36 +337,43 @@ static int osd_idc_find_and_init(const struct lu_env *env, * life-cycle. */ static struct lu_object *osd_object_alloc(const struct lu_env *env, - const struct lu_object_header *hdr, - struct lu_device *d) + const struct lu_object_header *hdr, + struct lu_device *d) { - struct osd_object *mo; + struct osd_object *mo; - OBD_ALLOC_PTR(mo); - if (mo != NULL) { - struct lu_object *l; + OBD_ALLOC_PTR(mo); + if (mo != NULL) { + struct lu_object *l; struct lu_object_header *h; + struct osd_device *o = osd_dev(d); l = &mo->oo_dt.do_lu; - h = &osd_dev(d)->od_scrub.os_scrub.os_obj_header; - if (unlikely(fid_is_zero(&h->loh_fid))) { - /* For the OI_scrub object during OSD device init. */ + if (unlikely(o->od_in_init)) { + OBD_ALLOC_PTR(h); + if (!h) { + OBD_FREE_PTR(mo); + return NULL; + } + lu_object_header_init(h); lu_object_init(l, h, d); lu_object_add_top(h, l); + mo->oo_header = h; } else { dt_object_init(&mo->oo_dt, NULL, d); + mo->oo_header = NULL; } + mo->oo_dt.do_ops = &osd_obj_ops; l->lo_ops = &osd_lu_obj_ops; init_rwsem(&mo->oo_sem); init_rwsem(&mo->oo_ext_idx_sem); spin_lock_init(&mo->oo_guard); INIT_LIST_HEAD(&mo->oo_xattr_list); - return l; - } else { - return NULL; - } + return l; + } + return NULL; } int osd_get_lma(struct osd_thread_info *info, struct inode *inode, @@ -391,10 +416,12 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, int rc; struct inode *inode = NULL; - /* if we look for an inode withing a running - * transaction, then we risk to deadlock */ - /* osd_dirent_check_repair() breaks this */ - /*LASSERT(current->journal_info == NULL);*/ + /* + * if we look for an inode withing a running + * transaction, then we risk to deadlock + * osd_dirent_check_repair() breaks this + */ + /* LASSERT(current->journal_info == NULL); */ inode = ldiskfs_iget(osd_sb(dev), id->oii_ino); if (IS_ERR(inode)) { @@ -408,14 +435,16 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, iput(inode); inode = ERR_PTR(-ESTALE); } else if (inode->i_nlink == 0) { - /* due to parallel readdir and unlink, - * we can have dead inode here. */ + /* + * due to parallel readdir and unlink, + * we can have dead inode here. + */ CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino); iput(inode); inode = ERR_PTR(-ESTALE); } else if (is_bad_inode(inode)) { CWARN("%s: bad inode: ino = %u\n", - osd_dev2name(dev), id->oii_ino); + osd_dev2name(dev), id->oii_ino); iput(inode); inode = ERR_PTR(-ENOENT); } else if ((rc = osd_attach_jinode(inode))) { @@ -426,11 +455,13 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, if (id->oii_gen == OSD_OII_NOGEN) osd_id_gen(id, inode->i_ino, inode->i_generation); - /* Do not update file c/mtime in ldiskfs. + /* + * Do not update file c/mtime in ldiskfs. * NB: we don't have any lock to protect this because we don't * have reference on osd_object now, but contention with * another lookup + attr_set can't happen in the tiny window - * between if (...) and set S_NOCMTIME. */ + * between if (...) and set S_NOCMTIME. + */ if (!(inode->i_flags & S_NOCMTIME)) inode->i_flags |= S_NOCMTIME; } @@ -483,7 +514,7 @@ int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd, } -static struct inode * +struct inode * osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev, struct osd_inode_id *id, struct lu_fid *fid) { @@ -518,11 +549,14 @@ static struct inode *osd_iget_check(struct osd_thread_info *info, { struct inode *inode; int rc = 0; + ENTRY; - /* The cached OI mapping is trustable. If we cannot locate the inode + /* + * The cached OI mapping is trustable. If we cannot locate the inode * via the cached OI mapping, then return the failure to the caller - * directly without further OI checking. */ + * directly without further OI checking. + */ again: inode = ldiskfs_iget(osd_sb(dev), id->oii_ino); @@ -579,7 +613,8 @@ check_oi: LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc); rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); - /* XXX: There are four possible cases: + /* + * XXX: There are four possible cases: * 1. rc = 0. * Backup/restore caused the OI invalid. * 2. rc = 0. @@ -598,16 +633,21 @@ check_oi: * if the OI files are consistent but may contain * stale OI mappings because of case 2, if iget() * returns -ENOENT or -ESTALE, then it should be - * the case 2. */ + * the case 2. + */ if (rc != 0) - /* If the OI mapping was in OI file before the + /* + * If the OI mapping was in OI file before the * osd_iget_check(), but now, it is disappear, * then it must be removed by race. That is a - * normal race case. */ + * normal race case. + */ GOTO(put, rc); - /* It is the OI scrub updated the OI mapping by race. - * The new OI mapping must be valid. */ + /* + * It is the OI scrub updated the OI mapping by race. + * The new OI mapping must be valid. + */ if (saved_ino != id->oii_ino || (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) { if (!IS_ERR(inode)) @@ -620,10 +660,12 @@ check_oi: if (IS_ERR(inode)) { if (dev->od_scrub.os_scrub.os_file.sf_flags & SF_INCONSISTENT) - /* It still can be the case 2, but we cannot + /* + * It still can be the case 2, but we cannot * distinguish it from the case 1. So return * -EREMCHG to block current operation until - * OI scrub rebuilt the OI mappings. */ + * OI scrub rebuilt the OI mappings. + */ rc = -EREMCHG; else rc = -ENOENT; @@ -639,11 +681,13 @@ check_oi: if (id->oii_gen == OSD_OII_NOGEN) osd_id_gen(id, inode->i_ino, inode->i_generation); - /* Do not update file c/mtime in ldiskfs. + /* + * Do not update file c/mtime in ldiskfs. * NB: we don't have any lock to protect this because we don't * have reference on osd_object now, but contention with * another lookup + attr_set can't happen in the tiny window - * between if (...) and set S_NOCMTIME. */ + * between if (...) and set S_NOCMTIME. + */ if (!(inode->i_flags & S_NOCMTIME)) inode->i_flags |= S_NOCMTIME; } @@ -669,9 +713,9 @@ put: int osd_get_idif(struct osd_thread_info *info, struct inode *inode, struct dentry *dentry, struct lu_fid *fid) { - struct filter_fid_old *ff = &info->oti_ff; - struct ost_id *ostid = &info->oti_ostid; - int rc; + struct filter_fid_old *ff = &info->oti_ff; + struct ost_id *ostid = &info->oti_ostid; + int rc; rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff)); if (rc == sizeof(*ff)) { @@ -680,7 +724,7 @@ int osd_get_idif(struct osd_thread_info *info, struct inode *inode, rc = ostid_set_id(ostid, le64_to_cpu(ff->ff_objid)); /* * XXX: use 0 as the index for compatibility, the caller will - * handle index related issues when necessary. + * handle index related issues when necessary. */ if (!rc) ostid_to_fid(fid, ostid, 0); @@ -698,7 +742,7 @@ static int osd_lma_self_repair(struct osd_thread_info *info, const struct lu_fid *fid, __u32 compat) { handle_t *jh; - int rc; + int rc; LASSERT(current->journal_info == NULL); @@ -721,15 +765,16 @@ static int osd_lma_self_repair(struct osd_thread_info *info, static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) { - struct osd_thread_info *info = osd_oti_get(env); - struct osd_device *osd = osd_obj2dev(obj); - struct lustre_ost_attrs *loa = &info->oti_ost_attrs; - struct lustre_mdt_attrs *lma = &loa->loa_lma; - struct inode *inode = obj->oo_inode; - struct dentry *dentry = &info->oti_obj_dentry; - struct lu_fid *fid = NULL; - const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu); - int rc; + struct osd_thread_info *info = osd_oti_get(env); + struct osd_device *osd = osd_obj2dev(obj); + struct lustre_ost_attrs *loa = &info->oti_ost_attrs; + struct lustre_mdt_attrs *lma = &loa->loa_lma; + struct inode *inode = obj->oo_inode; + struct dentry *dentry = &info->oti_obj_dentry; + struct lu_fid *fid = NULL; + const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu); + int rc; + ENTRY; rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, @@ -738,11 +783,13 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) fid = &lma->lma_self_fid; rc = osd_get_idif(info, inode, dentry, fid); if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) { - /* For the given OST-object, if it has neither LMA nor + /* + * For the given OST-object, if it has neither LMA nor * FID in XATTR_NAME_FID, then the given FID (which is * contained in the @obj, from client RPC for locating * the OST-object) is trusted. We use it to generate - * the LMA. */ + * the LMA. + */ osd_lma_self_repair(info, osd, inode, rfid, LMAC_FID_ON_OST); RETURN(0); @@ -780,10 +827,12 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) struct lu_fid *fid1 = &info->oti_fid3; __u32 idx = fid_idif_ost_idx(rfid); - /* For old IDIF, the OST index is not part of the IDIF, + /* + * For old IDIF, the OST index is not part of the IDIF, * Means that different OSTs may have the same IDIFs. * Under such case, we need to make some compatible - * check to make sure to trigger OI scrub properly. */ + * check to make sure to trigger OI scrub properly. + */ if (idx != 0 && fid_idif_ost_idx(fid) == 0) { /* Given @rfid is new, LMA is old. */ fid_to_ostid(fid, oi); @@ -807,11 +856,13 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) struct osd_check_lmv_buf { #ifdef HAVE_DIR_CONTEXT /* please keep it as first member */ - struct dir_context ctx; + struct dir_context ctx; #endif - struct osd_thread_info *oclb_info; - struct osd_device *oclb_dev; - struct osd_idmap_cache *oclb_oic; + struct osd_thread_info *oclb_info; + struct osd_device *oclb_dev; + struct osd_idmap_cache *oclb_oic; + int oclb_items; + bool oclb_found; }; /** @@ -828,7 +879,7 @@ static int osd_stripe_dir_filldir(struct dir_context *buf, static int osd_stripe_dir_filldir(void *buf, #endif const char *name, int namelen, - loff_t offset, __u64 ino, unsigned d_type) + loff_t offset, __u64 ino, unsigned int d_type) { struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf; struct osd_thread_info *oti = oclb->oclb_info; @@ -837,7 +888,8 @@ static int osd_stripe_dir_filldir(void *buf, struct osd_device *dev = oclb->oclb_dev; struct osd_idmap_cache *oic = oclb->oclb_oic; struct inode *inode; - int rc; + + oclb->oclb_items++; if (name[0] == '.') return 0; @@ -860,12 +912,14 @@ static int osd_stripe_dir_filldir(void *buf, oic->oic_fid = *fid; oic->oic_lid = *id; oic->oic_dev = dev; - rc = osd_oii_insert(dev, oic, true); + osd_oii_insert(dev, oic, true); + oclb->oclb_found = true; - return rc == 0 ? 1 : rc; + return 1; } -/* When lookup item under striped directory, we need to locate the master +/* + * When lookup item under striped directory, we need to locate the master * MDT-object of the striped directory firstly, then the client will send * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID * and the item's name. If the system is restored from MDT file level backup, @@ -898,7 +952,8 @@ static int osd_stripe_dir_filldir(void *buf, * slave MDT-objects information are stored as sub-directories with the name * "${FID}:${index}". Then when find the local slave MDT-object, its OI * mapping will be recorded. Then subsequent osd_fid_lookup() will know - * the correct OI mapping for the slave MDT-object. */ + * the correct OI mapping for the slave MDT-object. + */ static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev, struct inode *inode, struct osd_idmap_cache *oic) { @@ -913,9 +968,11 @@ static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev, #endif .oclb_info = oti, .oclb_dev = dev, - .oclb_oic = oic + .oclb_oic = oic, + .oclb_found = false, }; int rc = 0; + ENTRY; again: @@ -961,13 +1018,21 @@ again: filp->private_data = NULL; set_file_inode(filp, inode); + do { + oclb.oclb_items = 0; #ifdef HAVE_DIR_CONTEXT - oclb.ctx.pos = filp->f_pos; - rc = fops->iterate(filp, &oclb.ctx); - filp->f_pos = oclb.ctx.pos; + oclb.ctx.pos = filp->f_pos; +#ifdef HAVE_ITERATE_SHARED + rc = fops->iterate_shared(filp, &oclb.ctx); +#else + rc = fops->iterate(filp, &oclb.ctx); +#endif + filp->f_pos = oclb.ctx.pos; #else - rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir); + rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir); #endif + } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found && + filp->f_pos != LDISKFS_HTREE_EOF_64BIT); fops->release(inode, filp); out: @@ -1003,6 +1068,7 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, bool trusted = true; bool updated = false; bool checked = false; + ENTRY; LINVRNT(osd_invariant(obj)); @@ -1019,11 +1085,13 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT)) RETURN(-ENOENT); - /* For the object is created as locking anchor, or for the object to + /* + * For the object is created as locking anchor, or for the object to * be created on disk. No need to osd_oi_lookup() at here because FID * shouldn't never be re-used, if it's really a duplicate FID from * unexpected reason, we should be able to detect it later by calling - * do_create->osd_oi_insert(). */ + * do_create->osd_oi_insert(). + */ if (conf && conf->loc_flags & LOC_F_NEW) GOTO(out, result = 0); @@ -1041,8 +1109,10 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, goto iget; } - /* The OI mapping in the OI file can be updated by the OI scrub - * when we locate the inode via FID. So it may be not trustable. */ + /* + * The OI mapping in the OI file can be updated by the OI scrub + * when we locate the inode via FID. So it may be not trustable. + */ trusted = false; /* Search order: 3. OI files. */ @@ -1050,7 +1120,7 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, if (result == -ENOENT) { if (!(fid_is_norm(fid) || fid_is_igif(fid)) || fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) || - !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid), + !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)) GOTO(out, result = 0); @@ -1084,7 +1154,8 @@ iget: GOTO(out, result); trigger: - /* We still have chance to get the valid inode: for the + /* + * We still have chance to get the valid inode: for the * object which is referenced by remote name entry, the * object on the local MDT will be linked under the dir * of "/REMOTE_PARENT_DIR" with its FID string as name. @@ -1098,8 +1169,9 @@ trigger: * more, just lookup "/REMOTE_PARENT_DIR". Usually, it * only happened for the RPC from other MDT during the * OI scrub, or for the client side RPC with FID only, - * such as FID to path, or from old connected client. */ - if (!remote && !fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) { + * such as FID to path, or from old connected client. + */ + if (!remote) { rc1 = osd_lookup_in_remote_parent(info, dev, fid, id); if (!rc1) { remote = true; @@ -1154,8 +1226,11 @@ join: goto found; check_lma: - result = osd_check_lma(env, obj); checked = true; + if (unlikely(obj->oo_header)) + goto found; + + result = osd_check_lma(env, obj); if (!result) goto found; @@ -1169,30 +1244,38 @@ check_lma: saved_gen = inode->i_generation; if (unlikely(result == -ENODATA)) { - /* If the OI scrub updated the OI mapping by race, it - * must be valid. Trust the inode that has no LMA EA. */ + /* + * If the OI scrub updated the OI mapping by race, it + * must be valid. Trust the inode that has no LMA EA. + */ if (updated) goto found; result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); if (!result) { - /* The OI mapping is still there, the inode is still - * valid. It is just becaues the inode has no LMA EA. */ + /* + * The OI mapping is still there, the inode is still + * valid. It is just becaues the inode has no LMA EA. + */ if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) goto found; - /* It is the OI scrub updated the OI mapping by race. - * The new OI mapping must be valid. */ + /* + * It is the OI scrub updated the OI mapping by race. + * The new OI mapping must be valid. + */ trusted = true; updated = true; goto iget; } - /* "result == -ENOENT" means that the OI mappinghas been + /* + * "result == -ENOENT" means that the OI mappinghas been * removed by race, so the inode belongs to other object. * - * Others error can be returned directly. */ + * Others error can be returned directly. + */ if (result == -ENOENT) { LASSERT(trusted); @@ -1207,8 +1290,10 @@ check_lma: LASSERT(!updated); result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); - /* "result == -ENOENT" means the cached OI mapping has been removed - * from the OI file by race, above inode belongs to other object. */ + /* + * "result == -ENOENT" means the cached OI mapping has been removed + * from the OI file by race, above inode belongs to other object. + */ if (result == -ENOENT) { LASSERT(trusted); @@ -1224,8 +1309,10 @@ check_lma: goto trigger; } - /* It is the OI scrub updated the OI mapping by race. - * The new OI mapping must be valid. */ + /* + * It is the OI scrub updated the OI mapping by race. + * The new OI mapping must be valid. + */ trusted = true; updated = true; goto iget; @@ -1286,10 +1373,10 @@ out: */ static void osd_object_init0(struct osd_object *obj) { - LASSERT(obj->oo_inode != NULL); - obj->oo_dt.do_body_ops = &osd_body_ops; - obj->oo_dt.do_lu.lo_header->loh_attr |= - (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT)); + LASSERT(obj->oo_inode != NULL); + obj->oo_dt.do_body_ops = &osd_body_ops; + obj->oo_dt.do_lu.lo_header->loh_attr |= + (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT)); } /* @@ -1311,24 +1398,22 @@ static int osd_object_init(const struct lu_env *env, struct lu_object *l, } result = osd_fid_lookup(env, obj, lu_object_fid(l), conf); - if (unlikely(l->lo_header == - &osd_obj2dev(obj)->od_scrub.os_scrub.os_obj_header)) { - /* For the OI_scrub object during OSD device init. */ - l->lo_header->loh_attr |= LOHA_EXISTS; - obj->oo_dt.do_body_ops = &osd_body_ops; - } else { - obj->oo_dt.do_body_ops = &osd_body_ops_new; - } + obj->oo_dt.do_body_ops = &osd_body_ops_new; if (result == 0 && obj->oo_inode != NULL) { struct osd_thread_info *oti = osd_oti_get(env); struct lustre_ost_attrs *loa = &oti->oti_ost_attrs; osd_object_init0(obj); + if (unlikely(obj->oo_header)) + return 0; + result = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, loa); if (!result) { - /* Convert LMAI flags to lustre LMA flags - * and cache it to oo_lma_flags */ + /* + * Convert LMAI flags to lustre LMA flags + * and cache it to oo_lma_flags + */ obj->oo_lma_flags = lma_to_lustre_flags(loa->loa_lma.lma_incompat); } else if (result == -ENODATA) { @@ -1340,8 +1425,10 @@ static int osd_object_init(const struct lu_env *env, struct lu_object *l, return result; } -/* The first part of oxe_buf is xattr name, and is '\0' terminated. - * The left part is for value, binary mode. */ +/* + * The first part of oxe_buf is xattr name, and is '\0' terminated. + * The left part is for value, binary mode. + */ struct osd_xattr_entry { struct list_head oxe_list; size_t oxe_len; @@ -1358,6 +1445,7 @@ static int osd_oxc_get(struct osd_object *obj, const char *name, struct osd_xattr_entry *oxe = NULL; size_t namelen = strlen(name); int rc; + ENTRY; rcu_read_lock(); @@ -1478,21 +1566,19 @@ static void osd_oxc_fini(struct osd_object *obj) static void osd_object_free(const struct lu_env *env, struct lu_object *l) { struct osd_object *obj = osd_obj(l); - struct lu_object_header *h = NULL; + struct lu_object_header *h = obj->oo_header; LINVRNT(osd_invariant(obj)); - if (unlikely(l->lo_header == - &osd_obj2dev(obj)->od_scrub.os_scrub.os_obj_header)) - h = l->lo_header; - osd_oxc_fini(obj); dt_object_fini(&obj->oo_dt); if (obj->oo_hl_head != NULL) ldiskfs_htree_lock_head_free(obj->oo_hl_head); OBD_FREE_PTR(obj); - if (unlikely(h)) + if (unlikely(h)) { lu_object_header_fini(h); + OBD_FREE_PTR(h); + } } /* @@ -1501,17 +1587,17 @@ static void osd_object_free(const struct lu_env *env, struct lu_object *l) */ static void osd_index_fini(struct osd_object *o) { - struct iam_container *bag; + struct iam_container *bag; - if (o->oo_dir != NULL) { - bag = &o->oo_dir->od_container; - if (o->oo_inode != NULL) { - if (bag->ic_object == o->oo_inode) - iam_container_fini(bag); - } - OBD_FREE_PTR(o->oo_dir); - o->oo_dir = NULL; - } + if (o->oo_dir != NULL) { + bag = &o->oo_dir->od_container; + if (o->oo_inode != NULL) { + if (bag->ic_object == o->oo_inode) + iam_container_fini(bag); + } + OBD_FREE_PTR(o->oo_dir); + o->oo_dir = NULL; + } } /* @@ -1521,12 +1607,12 @@ static void osd_index_fini(struct osd_object *o) */ static int osd_inode_unlinked(const struct inode *inode) { - return inode->i_nlink == 0; + return inode->i_nlink == 0; } enum { - OSD_TXN_OI_DELETE_CREDITS = 20, - OSD_TXN_INODE_DELETE_CREDITS = 20 + OSD_TXN_OI_DELETE_CREDITS = 20, + OSD_TXN_INODE_DELETE_CREDITS = 20 }; /* @@ -1570,9 +1656,9 @@ static void __osd_th_check_slow(void *oth, struct osd_device *dev, if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) { CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n", - oth, now, alloced, started, closed); - libcfs_debug_dumpstack(NULL); - } + oth, now, alloced, started, closed); + libcfs_debug_dumpstack(NULL); + } } #define OSD_CHECK_SLOW_TH(oth, dev, expr) \ @@ -1608,19 +1694,19 @@ static int osd_param_is_not_sane(const struct osd_device *dev, * Concurrency: shouldn't matter. */ static void osd_trans_commit_cb(struct super_block *sb, - struct ldiskfs_journal_cb_entry *jcb, int error) + struct ldiskfs_journal_cb_entry *jcb, int error) { - struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb); - struct thandle *th = &oh->ot_super; - struct lu_device *lud = &th->th_dev->dd_lu_dev; - struct dt_txn_commit_cb *dcb, *tmp; + struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb); + struct thandle *th = &oh->ot_super; + struct lu_device *lud = &th->th_dev->dd_lu_dev; + struct dt_txn_commit_cb *dcb, *tmp; - LASSERT(oh->ot_handle == NULL); + LASSERT(oh->ot_handle == NULL); - if (error) - CERROR("transaction @0x%p commit error: %d\n", th, error); + if (error) + CERROR("transaction @0x%p commit error: %d\n", th, error); - dt_txn_hook_commit(th); + dt_txn_hook_commit(th); /* call per-transaction callbacks if any */ list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list, @@ -1633,8 +1719,8 @@ static void osd_trans_commit_cb(struct super_block *sb, } lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th); - lu_device_put(lud); - th->th_dev = NULL; + lu_device_put(lud); + th->th_dev = NULL; OBD_FREE_PTR(oh); } @@ -1647,10 +1733,11 @@ static void osd_trans_commit_cb(struct super_block *sb, static struct thandle *osd_trans_create(const struct lu_env *env, struct dt_device *d) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_iobuf *iobuf = &oti->oti_iobuf; - struct osd_thandle *oh; - struct thandle *th; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_iobuf *iobuf = &oti->oti_iobuf; + struct osd_thandle *oh; + struct thandle *th; + ENTRY; if (d->dd_rdonly) { @@ -1666,35 +1753,39 @@ static struct thandle *osd_trans_create(const struct lu_env *env, sb_start_write(osd_sb(osd_dt_dev(d))); - OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS); - if (oh != NULL) { - oh->ot_quota_trans = &oti->oti_quota_trans; - memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans)); - th = &oh->ot_super; - th->th_dev = d; - th->th_result = 0; - oh->ot_credits = 0; - INIT_LIST_HEAD(&oh->ot_commit_dcb_list); - INIT_LIST_HEAD(&oh->ot_stop_dcb_list); - osd_th_alloced(oh); - - memset(oti->oti_declare_ops, 0, - sizeof(oti->oti_declare_ops)); - memset(oti->oti_declare_ops_cred, 0, - sizeof(oti->oti_declare_ops_cred)); - memset(oti->oti_declare_ops_used, 0, - sizeof(oti->oti_declare_ops_used)); - } else { + OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS); + if (!oh) { sb_end_write(osd_sb(osd_dt_dev(d))); - th = ERR_PTR(-ENOMEM); + RETURN(ERR_PTR(-ENOMEM)); } + + oh->ot_quota_trans = &oti->oti_quota_trans; + memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans)); + th = &oh->ot_super; + th->th_dev = d; + th->th_result = 0; + oh->ot_credits = 0; + INIT_LIST_HEAD(&oh->ot_commit_dcb_list); + INIT_LIST_HEAD(&oh->ot_stop_dcb_list); + INIT_LIST_HEAD(&oh->ot_trunc_locks); + osd_th_alloced(oh); + + memset(oti->oti_declare_ops, 0, + sizeof(oti->oti_declare_ops)); + memset(oti->oti_declare_ops_cred, 0, + sizeof(oti->oti_declare_ops_cred)); + memset(oti->oti_declare_ops_used, 0, + sizeof(oti->oti_declare_ops_used)); + + oti->oti_ins_cache_depth++; + RETURN(th); } void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_thandle *oh; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_thandle *oh; oh = container_of0(th, struct osd_thandle, ot_super); LASSERT(oh != NULL); @@ -1745,30 +1836,32 @@ void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th) static int osd_trans_start(const struct lu_env *env, struct dt_device *d, struct thandle *th) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_device *dev = osd_dt_dev(d); - handle_t *jh; - struct osd_thandle *oh; - int rc; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_device *dev = osd_dt_dev(d); + handle_t *jh; + struct osd_thandle *oh; + int rc; - ENTRY; + ENTRY; - LASSERT(current->journal_info == NULL); + LASSERT(current->journal_info == NULL); - oh = container_of0(th, struct osd_thandle, ot_super); - LASSERT(oh != NULL); - LASSERT(oh->ot_handle == NULL); + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh != NULL); + LASSERT(oh->ot_handle == NULL); - rc = dt_txn_hook_start(env, d, th); - if (rc != 0) - GOTO(out, rc); + rc = dt_txn_hook_start(env, d, th); + if (rc != 0) + GOTO(out, rc); if (unlikely(osd_param_is_not_sane(dev, th))) { static unsigned long last_printed; static int last_credits; - /* don't make noise on a tiny testing systems - * actual credits misuse will be caught anyway */ + /* + * don't make noise on a tiny testing systems + * actual credits misuse will be caught anyway + */ if (last_credits != oh->ot_credits && time_after(jiffies, last_printed + msecs_to_jiffies(60 * MSEC_PER_SEC)) && @@ -1780,51 +1873,56 @@ static int osd_trans_start(const struct lu_env *env, struct dt_device *d, last_credits = oh->ot_credits; last_printed = jiffies; } - /* XXX Limit the credits to 'max_transaction_buffers', and + /* + * XXX Limit the credits to 'max_transaction_buffers', and * let the underlying filesystem to catch the error if * we really need so many credits. * * This should be removed when we can calculate the - * credits precisely. */ + * credits precisely. + */ oh->ot_credits = osd_transaction_size(dev); } else if (ldiskfs_track_declares_assert != 0) { - /* reserve few credits to prevent an assertion in JBD + /* + * reserve few credits to prevent an assertion in JBD * our debugging mechanism will be able to detected * overuse. this can help to debug single-update - * transactions */ + * transactions + */ oh->ot_credits += 10; if (unlikely(osd_param_is_not_sane(dev, th))) oh->ot_credits = osd_transaction_size(dev); } - /* - * XXX temporary stuff. Some abstraction layer should - * be used. - */ - jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits); - osd_th_started(oh); - if (!IS_ERR(jh)) { - oh->ot_handle = jh; - LASSERT(oti->oti_txns == 0); + /* + * XXX temporary stuff. Some abstraction layer should + * be used. + */ + jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits); + osd_th_started(oh); + if (!IS_ERR(jh)) { + oh->ot_handle = jh; + LASSERT(oti->oti_txns == 0); - lu_device_get(&d->dd_lu_dev); + lu_device_get(&d->dd_lu_dev); lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link, "osd-tx", th); - oti->oti_txns++; - rc = 0; - } else { - rc = PTR_ERR(jh); - } + oti->oti_txns++; + rc = 0; + } else { + rc = PTR_ERR(jh); + } out: - RETURN(rc); + RETURN(rc); } static int osd_seq_exists(const struct lu_env *env, struct osd_device *osd, u64 seq) { - struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range; - struct seq_server_site *ss = osd_seq_site(osd); - int rc; + struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range; + struct seq_server_site *ss = osd_seq_site(osd); + int rc; + ENTRY; LASSERT(ss != NULL); @@ -1843,8 +1941,8 @@ static int osd_seq_exists(const struct lu_env *env, static void osd_trans_stop_cb(struct osd_thandle *oth, int result) { - struct dt_txn_commit_cb *dcb; - struct dt_txn_commit_cb *tmp; + struct dt_txn_commit_cb *dcb; + struct dt_txn_commit_cb *tmp; /* call per-transaction stop callbacks if any */ list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list, @@ -1863,42 +1961,45 @@ static void osd_trans_stop_cb(struct osd_thandle *oth, int result) static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, struct thandle *th) { - int rc = 0, remove_agents = 0; - struct osd_thandle *oh; struct osd_thread_info *oti = osd_oti_get(env); - struct osd_iobuf *iobuf = &oti->oti_iobuf; - struct osd_device *osd = osd_dt_dev(th->th_dev); - struct qsd_instance *qsd = osd->od_quota_slave; - struct lquota_trans *qtrans; + struct osd_thandle *oh; + struct osd_iobuf *iobuf = &oti->oti_iobuf; + struct osd_device *osd = osd_dt_dev(th->th_dev); + struct qsd_instance *qsd = osd->od_quota_slave; + struct lquota_trans *qtrans; + struct list_head truncates = LIST_HEAD_INIT(truncates); + int rc = 0, remove_agents = 0; + ENTRY; oh = container_of0(th, struct osd_thandle, ot_super); - /* reset OI cache for safety */ - oti->oti_ins_cache_used = 0; - remove_agents = oh->ot_remove_agents; qtrans = oh->ot_quota_trans; oh->ot_quota_trans = NULL; + /* move locks to local list, stop tx, execute truncates */ + list_splice(&oh->ot_trunc_locks, &truncates); + if (oh->ot_handle != NULL) { int rc2; - handle_t *hdl = oh->ot_handle; - /* - * add commit callback - * notice we don't do this in osd_trans_start() - * as underlying transaction can change during truncate - */ - ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb, - &oh->ot_jcb); + handle_t *hdl = oh->ot_handle; + + /* + * add commit callback + * notice we don't do this in osd_trans_start() + * as underlying transaction can change during truncate + */ + ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb, + &oh->ot_jcb); - LASSERT(oti->oti_txns == 1); - oti->oti_txns--; + LASSERT(oti->oti_txns == 1); + oti->oti_txns--; - rc = dt_txn_hook_stop(env, th); - if (rc != 0) + rc = dt_txn_hook_stop(env, th); + if (rc != 0) CERROR("%s: failed in transaction hook: rc = %d\n", osd_name(osd), rc); @@ -1913,15 +2014,20 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, osd_name(osd), rc2); if (!rc) rc = rc2; + + osd_process_truncates(&truncates); } else { osd_trans_stop_cb(oh, th->th_result); OBD_FREE_PTR(oh); } + osd_trunc_unlock_all(&truncates); + /* inform the quota slave device that the transaction is stopping */ qsd_op_end(env, qsd, qtrans); - /* as we want IO to journal and data IO be concurrent, we don't block + /* + * as we want IO to journal and data IO be concurrent, we don't block * awaiting data IO completion in osd_do_bio(), instead we wait here * once transaction is submitted to the journal. all reqular requests * don't do direct IO (except read/write), thus this wait_event becomes @@ -1939,6 +2045,11 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, if (unlikely(remove_agents != 0)) osd_process_scheduled_agent_removals(env, osd); + oti->oti_ins_cache_depth--; + /* reset OI cache for safety */ + if (oti->oti_ins_cache_depth == 0) + oti->oti_ins_cache_used = 0; + sb_end_write(osd_sb(osd)); RETURN(rc); @@ -1968,27 +2079,26 @@ static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb) */ static void osd_object_delete(const struct lu_env *env, struct lu_object *l) { - struct osd_object *obj = osd_obj(l); - struct inode *inode = obj->oo_inode; + struct osd_object *obj = osd_obj(l); + struct inode *inode = obj->oo_inode; - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); - /* - * If object is unlinked remove fid->ino mapping from object index. - */ + /* + * If object is unlinked remove fid->ino mapping from object index. + */ - osd_index_fini(obj); - if (inode != NULL) { + osd_index_fini(obj); + if (inode != NULL) { struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave; qid_t uid = i_uid_read(inode); qid_t gid = i_gid_read(inode); - iput(inode); - obj->oo_inode = NULL; - - if (qsd != NULL) { - struct osd_thread_info *info = osd_oti_get(env); - struct lquota_id_info *qi = &info->oti_qi; + obj->oo_inode = NULL; + iput(inode); + if (!obj->oo_header && qsd) { + struct osd_thread_info *info = osd_oti_get(env); + struct lquota_id_info *qi = &info->oti_qi; /* Release granted quota to master if necessary */ qi->lqi_id.qid_uid = uid; @@ -2000,67 +2110,70 @@ static void osd_object_delete(const struct lu_env *env, struct lu_object *l) qi->lqi_id.qid_uid = i_projid_read(inode); qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA); } - } + } } /* * Concurrency: ->loo_object_release() is called under site spin-lock. */ static void osd_object_release(const struct lu_env *env, - struct lu_object *l) + struct lu_object *l) { struct osd_object *o = osd_obj(l); - /* nobody should be releasing a non-destroyed object with nlink=0 + + /* + * nobody should be releasing a non-destroyed object with nlink=0 * the API allows this, but ldiskfs doesn't like and then report - * this inode as deleted */ - if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0)) - LBUG(); + * this inode as deleted + */ + LASSERT(!(o->oo_destroyed == 0 && o->oo_inode && + o->oo_inode->i_nlink == 0)); } /* * Concurrency: shouldn't matter. */ static int osd_object_print(const struct lu_env *env, void *cookie, - lu_printer_t p, const struct lu_object *l) + lu_printer_t p, const struct lu_object *l) { - struct osd_object *o = osd_obj(l); - struct iam_descr *d; + struct osd_object *o = osd_obj(l); + struct iam_descr *d; - if (o->oo_dir != NULL) - d = o->oo_dir->od_container.ic_descr; - else - d = NULL; + if (o->oo_dir != NULL) + d = o->oo_dir->od_container.ic_descr; + else + d = NULL; return (*p)(env, cookie, LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]", - o, o->oo_inode, - o->oo_inode ? o->oo_inode->i_ino : 0UL, - o->oo_inode ? o->oo_inode->i_generation : 0, - d ? d->id_ops->id_name : "plain"); + o, o->oo_inode, + o->oo_inode ? o->oo_inode->i_ino : 0UL, + o->oo_inode ? o->oo_inode->i_generation : 0, + d ? d->id_ops->id_name : "plain"); } /* * Concurrency: shouldn't matter. */ int osd_statfs(const struct lu_env *env, struct dt_device *d, - struct obd_statfs *sfs) + struct obd_statfs *sfs) { - struct osd_device *osd = osd_dt_dev(d); - struct super_block *sb = osd_sb(osd); - struct kstatfs *ksfs; - __u64 reserved; - int result = 0; + struct osd_device *osd = osd_dt_dev(d); + struct super_block *sb = osd_sb(osd); + struct kstatfs *ksfs; + __u64 reserved; + int result = 0; if (unlikely(osd->od_mnt == NULL)) return -EINPROGRESS; - /* osd_lproc.c call this without env, allocate ksfs for that case */ - if (unlikely(env == NULL)) { - OBD_ALLOC_PTR(ksfs); - if (ksfs == NULL) - return -ENOMEM; - } else { - ksfs = &osd_oti_get(env)->oti_ksfs; - } + /* osd_lproc.c call this without env, allocate ksfs for that case */ + if (unlikely(env == NULL)) { + OBD_ALLOC_PTR(ksfs); + if (ksfs == NULL) + return -ENOMEM; + } else { + ksfs = &osd_oti_get(env)->oti_ksfs; + } result = sb->s_op->statfs(sb->s_root, ksfs); if (result) @@ -2069,8 +2182,7 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d, statfs_pack(sfs, ksfs); if (unlikely(sb->s_flags & MS_RDONLY)) sfs->os_state |= OS_STATE_READONLY; - if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, - LDISKFS_FEATURE_INCOMPAT_EXTENTS)) + if (ldiskfs_has_feature_extents(sb)) sfs->os_maxbytes = sb->s_maxbytes; else sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes; @@ -2113,52 +2225,113 @@ out: * Concurrency: doesn't access mutable data. */ static void osd_conf_get(const struct lu_env *env, - const struct dt_device *dev, - struct dt_device_param *param) + const struct dt_device *dev, + struct dt_device_param *param) { - struct super_block *sb = osd_sb(osd_dt_dev(dev)); - int ea_overhead; + struct osd_device *d = osd_dt_dev(dev); + struct super_block *sb = osd_sb(d); + struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev); + const char *name; + int ea_overhead; - /* - * XXX should be taken from not-yet-existing fs abstraction layer. - */ - param->ddp_max_name_len = LDISKFS_NAME_LEN; - param->ddp_max_nlink = LDISKFS_LINK_MAX; + /* + * XXX should be taken from not-yet-existing fs abstraction layer. + */ + param->ddp_max_name_len = LDISKFS_NAME_LEN; + param->ddp_max_nlink = LDISKFS_LINK_MAX; param->ddp_symlink_max = sb->s_blocksize; - param->ddp_mount_type = LDD_MT_LDISKFS; - if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS)) + param->ddp_mount_type = LDD_MT_LDISKFS; + if (ldiskfs_has_feature_extents(sb)) param->ddp_maxbytes = sb->s_maxbytes; else param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes; - /* inode are statically allocated, so per-inode space consumption - * is the space consumed by the directory entry */ + /* + * inode are statically allocated, so per-inode space consumption + * is the space consumed by the directory entry + */ param->ddp_inodespace = PER_OBJ_USAGE; - /* EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks + /* + * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks * = 128MB) which is unlikely to be hit in real life. Report a smaller * maximum length to not under count the actual number of extents - * needed for writing a file. */ + * needed for writing a file. + */ param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2; /* worst-case extent insertion metadata overhead */ param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb); - param->ddp_mntopts = 0; - if (test_opt(sb, XATTR_USER)) - param->ddp_mntopts |= MNTOPT_USERXATTR; - if (test_opt(sb, POSIX_ACL)) - param->ddp_mntopts |= MNTOPT_ACL; + param->ddp_mntopts = 0; + if (test_opt(sb, XATTR_USER)) + param->ddp_mntopts |= MNTOPT_USERXATTR; + if (test_opt(sb, POSIX_ACL)) + param->ddp_mntopts |= MNTOPT_ACL; - /* LOD might calculate the max stripe count based on max_ea_size, + /* + * LOD might calculate the max stripe count based on max_ea_size, * so we need take account in the overhead as well, - * xattr_header + magic + xattr_entry_head */ + * xattr_header + magic + xattr_entry_head + */ ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) + LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN); #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE) - if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE)) + if (ldiskfs_has_feature_ea_inode(sb)) param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE - ea_overhead; else #endif param->ddp_max_ea_size = sb->s_blocksize - ea_overhead; + + if (param->ddp_max_ea_size > OSD_MAX_EA_SIZE) + param->ddp_max_ea_size = OSD_MAX_EA_SIZE; + + /* + * Preferred RPC size for efficient disk IO. 4MB shows good + * all-around performance for ldiskfs, but use bigalloc chunk size + * by default if larger. + */ +#if defined(LDISKFS_CLUSTER_SIZE) + if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE) + param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb); + else +#endif + param->ddp_brw_size = DT_DEF_BRW_SIZE; + + param->ddp_t10_cksum_type = 0; + if (bi) { + unsigned short interval = blk_integrity_interval(bi); + name = blk_integrity_name(bi); + /* + * Expected values: + * T10-DIF-TYPE1-CRC + * T10-DIF-TYPE3-CRC + * T10-DIF-TYPE1-IP + * T10-DIF-TYPE3-IP + */ + if (strncmp(name, "T10-DIF-TYPE", + sizeof("T10-DIF-TYPE") - 1) == 0) { + /* also skip "1/3-" at end */ + const int type_off = sizeof("T10-DIF-TYPE."); + + if (interval != 512 && interval != 4096) + CERROR("%s: unsupported T10PI sector size %u\n", + d->od_svname, interval); + else if (strcmp(name + type_off, "CRC") == 0) + param->ddp_t10_cksum_type = interval == 512 ? + OBD_CKSUM_T10CRC512 : + OBD_CKSUM_T10CRC4K; + else if (strcmp(name + type_off, "IP") == 0) + param->ddp_t10_cksum_type = interval == 512 ? + OBD_CKSUM_T10IP512 : + OBD_CKSUM_T10IP4K; + else + CERROR("%s: unsupported checksum type of " + "T10PI type '%s'", + d->od_svname, name); + } else { + CERROR("%s: unsupported T10PI type '%s'", + d->od_svname, name); + } + } } /* @@ -2167,10 +2340,12 @@ static void osd_conf_get(const struct lu_env *env, static int osd_sync(const struct lu_env *env, struct dt_device *d) { int rc; + struct super_block *s = osd_sb(osd_dt_dev(d)); + ENTRY; - CDEBUG(D_CACHE, "%s: syncing OSD\n", osd_dt_dev(d)->od_svname); - - rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d))); + down_read(&s->s_umount); + rc = s->s_op->sync_fs(s, 1); + up_read(&s->s_umount); CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname, rc); @@ -2191,13 +2366,19 @@ static int osd_sync(const struct lu_env *env, struct dt_device *d) * \see dt_device_operations */ static int osd_commit_async(const struct lu_env *env, - struct dt_device *d) + struct dt_device *d) { - struct super_block *s = osd_sb(osd_dt_dev(d)); - ENTRY; + struct super_block *s = osd_sb(osd_dt_dev(d)); + int rc; + + ENTRY; CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname); - RETURN(s->s_op->sync_fs(s, 0)); + down_read(&s->s_umount); + rc = s->s_op->sync_fs(s, 0); + up_read(&s->s_umount); + + RETURN(rc); } /* Our own copy of the set readonly functions if present, or NU if not. */ @@ -2213,6 +2394,7 @@ static int osd_ro(const struct lu_env *env, struct dt_device *d) struct super_block *sb = osd_sb(osd_dt_dev(d)); struct block_device *dev = sb->s_bdev; int rc = -EOPNOTSUPP; + ENTRY; if (priv_dev_set_rdonly) { @@ -2307,20 +2489,20 @@ const int osd_dto_credits_noquota[DTO_NR] = { }; static const struct dt_device_operations osd_dt_ops = { - .dt_root_get = osd_root_get, - .dt_statfs = osd_statfs, - .dt_trans_create = osd_trans_create, - .dt_trans_start = osd_trans_start, - .dt_trans_stop = osd_trans_stop, - .dt_trans_cb_add = osd_trans_cb_add, - .dt_conf_get = osd_conf_get, - .dt_sync = osd_sync, - .dt_ro = osd_ro, - .dt_commit_async = osd_commit_async, + .dt_root_get = osd_root_get, + .dt_statfs = osd_statfs, + .dt_trans_create = osd_trans_create, + .dt_trans_start = osd_trans_start, + .dt_trans_stop = osd_trans_stop, + .dt_trans_cb_add = osd_trans_cb_add, + .dt_conf_get = osd_conf_get, + .dt_sync = osd_sync, + .dt_ro = osd_ro, + .dt_commit_async = osd_commit_async, }; static void osd_read_lock(const struct lu_env *env, struct dt_object *dt, - unsigned role) + unsigned int role) { struct osd_object *obj = osd_dt_obj(dt); struct osd_thread_info *oti = osd_oti_get(env); @@ -2335,7 +2517,7 @@ static void osd_read_lock(const struct lu_env *env, struct dt_object *dt, } static void osd_write_lock(const struct lu_env *env, struct dt_object *dt, - unsigned role) + unsigned int role) { struct osd_object *obj = osd_dt_obj(dt); struct osd_thread_info *oti = osd_oti_get(env); @@ -2388,8 +2570,8 @@ static int osd_write_locked(const struct lu_env *env, struct dt_object *dt) static struct timespec *osd_inode_time(const struct lu_env *env, struct inode *inode, __u64 seconds) { - struct osd_thread_info *oti = osd_oti_get(env); - struct timespec *t = &oti->oti_time; + struct osd_thread_info *oti = osd_oti_get(env); + struct timespec *t = &oti->oti_time; t->tv_sec = seconds; t->tv_nsec = 0; @@ -2455,7 +2637,7 @@ static int osd_declare_attr_qid(const struct lu_env *env, struct osd_object *obj, struct osd_thandle *oh, long long bspace, qid_t old_id, qid_t new_id, bool enforce, - unsigned type) + unsigned int type, bool ignore_edquot) { int rc; struct osd_thread_info *info = osd_oti_get(env); @@ -2470,16 +2652,16 @@ static int osd_declare_attr_qid(const struct lu_env *env, qi->lqi_space = 1; /* Reserve credits for the new id */ rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL); - if (rc == -EDQUOT || rc == -EINPROGRESS) + if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS)) rc = 0; if (rc) RETURN(rc); /* and one less inode for the current id */ qi->lqi_id.qid_uid = old_id; - qi->lqi_space = -1; + qi->lqi_space = -1; rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); - if (rc == -EDQUOT || rc == -EINPROGRESS) + if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS)) rc = 0; if (rc) RETURN(rc); @@ -2495,7 +2677,7 @@ static int osd_declare_attr_qid(const struct lu_env *env, * to save credit reservation. */ rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); - if (rc == -EDQUOT || rc == -EINPROGRESS) + if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS)) rc = 0; if (rc) RETURN(rc); @@ -2504,24 +2686,25 @@ static int osd_declare_attr_qid(const struct lu_env *env, qi->lqi_id.qid_uid = old_id; qi->lqi_space = -bspace; rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); - if (rc == -EDQUOT || rc == -EINPROGRESS) + if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS)) rc = 0; RETURN(rc); } static int osd_declare_attr_set(const struct lu_env *env, - struct dt_object *dt, - const struct lu_attr *attr, - struct thandle *handle) + struct dt_object *dt, + const struct lu_attr *attr, + struct thandle *handle) { - struct osd_thandle *oh; - struct osd_object *obj; - qid_t uid; - qid_t gid; - long long bspace; - int rc = 0; - bool enforce; + struct osd_thandle *oh; + struct osd_object *obj; + qid_t uid; + qid_t gid; + long long bspace; + int rc = 0; + bool enforce; + ENTRY; LASSERT(dt != NULL); @@ -2542,22 +2725,32 @@ static int osd_declare_attr_set(const struct lu_env *env, if (attr == NULL || obj->oo_inode == NULL) RETURN(rc); - bspace = obj->oo_inode->i_blocks; - bspace <<= obj->oo_inode->i_sb->s_blocksize_bits; + bspace = obj->oo_inode->i_blocks << 9; bspace = toqb(bspace); - /* Changing ownership is always preformed by super user, it should not - * fail with EDQUOT. + /* + * Changing ownership is always preformed by super user, it should not + * fail with EDQUOT unless required explicitly. * * We still need to call the osd_declare_qid() to calculate the journal * credits for updating quota accounting files and to trigger quota - * space adjustment once the operation is completed.*/ + * space adjustment once the operation is completed. + */ if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) { + bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL); + + if (!ignore_edquot) + CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u" + "(the quota space is %lld)\n", + obj->oo_inode->i_sb->s_id, attr->la_uid, + attr->la_gid, bspace); + /* USERQUOTA */ uid = i_uid_read(obj->oo_inode); enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid); rc = osd_declare_attr_qid(env, obj, oh, bspace, uid, - attr->la_uid, enforce, USRQUOTA); + attr->la_uid, enforce, USRQUOTA, + true); if (rc) RETURN(rc); @@ -2565,7 +2758,8 @@ static int osd_declare_attr_set(const struct lu_env *env, enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid); rc = osd_declare_attr_qid(env, obj, oh, bspace, i_gid_read(obj->oo_inode), - attr->la_gid, enforce, GRPQUOTA); + attr->la_gid, enforce, GRPQUOTA, + ignore_edquot); if (rc) RETURN(rc); @@ -2573,11 +2767,12 @@ static int osd_declare_attr_set(const struct lu_env *env, #ifdef HAVE_PROJECT_QUOTA if (attr->la_valid & LA_PROJID) { __u32 projid = i_projid_read(obj->oo_inode); + enforce = (attr->la_valid & LA_PROJID) && (attr->la_projid != projid); rc = osd_declare_attr_qid(env, obj, oh, bspace, (qid_t)projid, (qid_t)attr->la_projid, - enforce, PRJQUOTA); + enforce, PRJQUOTA, true); if (rc) RETURN(rc); } @@ -2610,8 +2805,10 @@ static int osd_inode_setattr(const struct lu_env *env, spin_unlock(&inode->i_lock); } - /* OSD should not change "i_blocks" which is used by quota. - * "i_blocks" should be changed by ldiskfs only. */ + /* + * OSD should not change "i_blocks" which is used by quota. + * "i_blocks" should be changed by ldiskfs only. + */ if (bits & LA_MODE) inode->i_mode = (inode->i_mode & S_IFMT) | (attr->la_mode & ~S_IFMT); @@ -2643,13 +2840,65 @@ static int osd_inode_setattr(const struct lu_env *env, return 0; } +#ifdef HAVE_PROJECT_QUOTA +static int osd_transfer_project(struct inode *inode, __u32 projid) +{ + struct super_block *sb = inode->i_sb; + struct ldiskfs_inode_info *ei = LDISKFS_I(inode); + int err; + kprojid_t kprojid; + struct ldiskfs_iloc iloc; + struct ldiskfs_inode *raw_inode; + struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { }; + + if (!ldiskfs_has_feature_project(sb)) { + LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid) + == LDISKFS_DEF_PROJID); + if (projid != LDISKFS_DEF_PROJID) + return -EOPNOTSUPP; + else + return 0; + } + + if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE) + return -EOPNOTSUPP; + + kprojid = make_kprojid(&init_user_ns, (projid_t)projid); + if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid)) + return 0; + + err = ldiskfs_get_inode_loc(inode, &iloc); + if (err) + return err; + + raw_inode = ldiskfs_raw_inode(&iloc); + if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) { + err = -EOVERFLOW; + brelse(iloc.bh); + return err; + } + brelse(iloc.bh); + + dquot_initialize(inode); + transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid)); + if (transfer_to[PRJQUOTA]) { + err = __dquot_transfer(inode, transfer_to); + dqput(transfer_to[PRJQUOTA]); + if (err) + return err; + } + + return err; +} +#endif + static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) { int rc; if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) || (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) { - struct iattr iattr; + struct iattr iattr; ll_vfs_dq_init(inode); iattr.ia_valid = 0; @@ -2673,7 +2922,7 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) if (attr->la_valid & LA_PROJID && attr->la_projid != i_projid_read(inode)) { #ifdef HAVE_PROJECT_QUOTA - rc = ldiskfs_transfer_project(inode, attr->la_projid); + rc = osd_transfer_project(inode, attr->la_projid); #else rc = -ENOTSUPP; #endif @@ -2693,7 +2942,7 @@ static int osd_attr_set(const struct lu_env *env, struct thandle *handle) { struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode; + struct inode *inode; int rc; if (!dt_object_exists(dt)) @@ -2707,14 +2956,14 @@ static int osd_attr_set(const struct lu_env *env, if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) && !osd_obj2dev(obj)->od_is_ost) { - struct osd_thread_info *oti = osd_oti_get(env); - const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu); - struct lu_fid *fid1 = &oti->oti_fid; - struct osd_inode_id *id = &oti->oti_id; - struct iam_path_descr *ipd; - struct iam_container *bag; - struct osd_thandle *oh; - int rc; + struct osd_thread_info *oti = osd_oti_get(env); + const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu); + struct lu_fid *fid1 = &oti->oti_fid; + struct osd_inode_id *id = &oti->oti_id; + struct iam_path_descr *ipd; + struct iam_container *bag; + struct osd_thandle *oh; + int rc; fid_cpu_to_be(fid1, fid0); memset(id, 1, sizeof(*id)); @@ -2725,13 +2974,14 @@ static int osd_attr_set(const struct lu_env *env, RETURN(-ENOMEM); oh = container_of0(handle, struct osd_thandle, ot_super); - rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1, + rc = iam_update(oh->ot_handle, bag, + (const struct iam_key *)fid1, (const struct iam_rec *)id, ipd); osd_ipd_put(env, bag, ipd); return(rc > 0 ? 0 : rc); } - inode = obj->oo_inode; + inode = obj->oo_inode; rc = osd_quota_transfer(inode, attr); if (rc) @@ -2780,70 +3030,73 @@ static int osd_attr_set(const struct lu_env *env, out: osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET); - return rc; + return rc; } static struct dentry *osd_child_dentry_get(const struct lu_env *env, struct osd_object *obj, const char *name, const int namelen) { - return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen); + return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen); } static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj, umode_t mode, struct dt_allocation_hint *hint, struct thandle *th) { - int result; - struct osd_device *osd = osd_obj2dev(obj); - struct osd_thandle *oth; - struct dt_object *parent = NULL; - struct inode *inode; + int result; + struct osd_device *osd = osd_obj2dev(obj); + struct osd_thandle *oth; + struct dt_object *parent = NULL; + struct inode *inode; - LINVRNT(osd_invariant(obj)); - LASSERT(obj->oo_inode == NULL); - LASSERT(obj->oo_hl_head == NULL); + LINVRNT(osd_invariant(obj)); + LASSERT(obj->oo_inode == NULL); + LASSERT(obj->oo_hl_head == NULL); - if (S_ISDIR(mode) && ldiskfs_pdo) { - obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF); - if (obj->oo_hl_head == NULL) - return -ENOMEM; - } + if (S_ISDIR(mode) && ldiskfs_pdo) { + obj->oo_hl_head = + ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF); + if (obj->oo_hl_head == NULL) + return -ENOMEM; + } - oth = container_of(th, struct osd_thandle, ot_super); - LASSERT(oth->ot_handle->h_transaction != NULL); + oth = container_of(th, struct osd_thandle, ot_super); + LASSERT(oth->ot_handle->h_transaction != NULL); if (hint != NULL && hint->dah_parent != NULL && !dt_object_remote(hint->dah_parent)) parent = hint->dah_parent; - inode = ldiskfs_create_inode(oth->ot_handle, - parent ? osd_dt_obj(parent)->oo_inode : - osd_sb(osd)->s_root->d_inode, - mode); - if (!IS_ERR(inode)) { + inode = ldiskfs_create_inode(oth->ot_handle, + parent ? osd_dt_obj(parent)->oo_inode : + osd_sb(osd)->s_root->d_inode, + mode); + if (!IS_ERR(inode)) { /* Do not update file c/mtime in ldiskfs. */ inode->i_flags |= S_NOCMTIME; - /* For new created object, it must be consistent, - * and it is unnecessary to scrub against it. */ + /* + * For new created object, it must be consistent, + * and it is unnecessary to scrub against it. + */ ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB); - obj->oo_inode = inode; - result = 0; - } else { - if (obj->oo_hl_head != NULL) { - ldiskfs_htree_lock_head_free(obj->oo_hl_head); - obj->oo_hl_head = NULL; - } - result = PTR_ERR(inode); - } - LINVRNT(osd_invariant(obj)); - return result; + obj->oo_inode = inode; + result = 0; + } else { + if (obj->oo_hl_head != NULL) { + ldiskfs_htree_lock_head_free(obj->oo_hl_head); + obj->oo_hl_head = NULL; + } + result = PTR_ERR(inode); + } + LINVRNT(osd_invariant(obj)); + return result; } enum { - OSD_NAME_LEN = 255 + OSD_NAME_LEN = 255 }; static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj, @@ -2866,62 +3119,61 @@ static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj, } static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj, - struct lu_attr *attr, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *th) -{ - int result; - struct osd_thandle *oth; - const struct dt_index_features *feat = dof->u.dof_idx.di_feat; - - __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX)); - - LASSERT(S_ISREG(attr->la_mode)); - - oth = container_of(th, struct osd_thandle, ot_super); - LASSERT(oth->ot_handle->h_transaction != NULL); - - result = osd_mkfile(info, obj, mode, hint, th); - if (result == 0) { - LASSERT(obj->oo_inode != NULL); - if (feat->dif_flags & DT_IND_VARKEY) - result = iam_lvar_create(obj->oo_inode, - feat->dif_keysize_max, - feat->dif_ptrsize, - feat->dif_recsize_max, - oth->ot_handle); - else - result = iam_lfix_create(obj->oo_inode, - feat->dif_keysize_max, - feat->dif_ptrsize, - feat->dif_recsize_max, - oth->ot_handle); - - } - return result; + struct lu_attr *attr, + struct dt_allocation_hint *hint, + struct dt_object_format *dof, + struct thandle *th) +{ + int result; + struct osd_thandle *oth; + const struct dt_index_features *feat = dof->u.dof_idx.di_feat; + + __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX)); + + LASSERT(S_ISREG(attr->la_mode)); + + oth = container_of(th, struct osd_thandle, ot_super); + LASSERT(oth->ot_handle->h_transaction != NULL); + + result = osd_mkfile(info, obj, mode, hint, th); + if (result == 0) { + LASSERT(obj->oo_inode != NULL); + if (feat->dif_flags & DT_IND_VARKEY) + result = iam_lvar_create(obj->oo_inode, + feat->dif_keysize_max, + feat->dif_ptrsize, + feat->dif_recsize_max, + oth->ot_handle); + else + result = iam_lfix_create(obj->oo_inode, + feat->dif_keysize_max, + feat->dif_ptrsize, + feat->dif_recsize_max, + oth->ot_handle); + } + return result; } static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj, - struct lu_attr *attr, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *th) + struct lu_attr *attr, + struct dt_allocation_hint *hint, + struct dt_object_format *dof, + struct thandle *th) { - LASSERT(S_ISREG(attr->la_mode)); - return osd_mkfile(info, obj, (attr->la_mode & - (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th); + LASSERT(S_ISREG(attr->la_mode)); + return osd_mkfile(info, obj, (attr->la_mode & + (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th); } static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj, - struct lu_attr *attr, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *th) + struct lu_attr *attr, + struct dt_allocation_hint *hint, + struct dt_object_format *dof, + struct thandle *th) { - LASSERT(S_ISLNK(attr->la_mode)); - return osd_mkfile(info, obj, (attr->la_mode & - (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th); + LASSERT(S_ISLNK(attr->la_mode)); + return osd_mkfile(info, obj, (attr->la_mode & + (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th); } static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj, @@ -2935,55 +3187,55 @@ static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj, LINVRNT(osd_invariant(obj)); LASSERT(obj->oo_inode == NULL); - LASSERT(S_ISCHR(mode) || S_ISBLK(mode) || - S_ISFIFO(mode) || S_ISSOCK(mode)); + LASSERT(S_ISCHR(mode) || S_ISBLK(mode) || + S_ISFIFO(mode) || S_ISSOCK(mode)); - result = osd_mkfile(info, obj, mode, hint, th); - if (result == 0) { - LASSERT(obj->oo_inode != NULL); + result = osd_mkfile(info, obj, mode, hint, th); + if (result == 0) { + LASSERT(obj->oo_inode != NULL); /* * This inode should be marked dirty for i_rdev. Currently * that is done in the osd_attr_init(). */ init_special_inode(obj->oo_inode, obj->oo_inode->i_mode, attr->la_rdev); - } - LINVRNT(osd_invariant(obj)); - return result; + } + LINVRNT(osd_invariant(obj)); + return result; } typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *, - struct lu_attr *, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *); + struct lu_attr *, + struct dt_allocation_hint *hint, + struct dt_object_format *dof, + struct thandle *); static osd_obj_type_f osd_create_type_f(enum dt_format_type type) { - osd_obj_type_f result; - - switch (type) { - case DFT_DIR: - result = osd_mkdir; - break; - case DFT_REGULAR: - result = osd_mkreg; - break; - case DFT_SYM: - result = osd_mksym; - break; - case DFT_NODE: - result = osd_mknod; - break; - case DFT_INDEX: - result = osd_mk_index; - break; - - default: - LBUG(); - break; - } - return result; + osd_obj_type_f result; + + switch (type) { + case DFT_DIR: + result = osd_mkdir; + break; + case DFT_REGULAR: + result = osd_mkreg; + break; + case DFT_SYM: + result = osd_mksym; + break; + case DFT_NODE: + result = osd_mknod; + break; + case DFT_INDEX: + result = osd_mk_index; + break; + + default: + LBUG(); + break; + } + return result; } @@ -2991,14 +3243,15 @@ static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah, struct dt_object *parent, struct dt_object *child, umode_t child_mode) { - LASSERT(ah); + LASSERT(ah); - ah->dah_parent = parent; - ah->dah_mode = child_mode; + ah->dah_parent = parent; + ah->dah_mode = child_mode; if (parent != NULL && !dt_object_remote(parent)) { /* will help to find FID->ino at dt_insert("..") */ struct osd_object *pobj = osd_dt_obj(parent); + osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj); } } @@ -3006,20 +3259,20 @@ static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah, static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj, struct lu_attr *attr, struct dt_object_format *dof) { - struct inode *inode = obj->oo_inode; - __u64 valid = attr->la_valid; - int result; + struct inode *inode = obj->oo_inode; + __u64 valid = attr->la_valid; + int result; attr->la_valid &= ~(LA_TYPE | LA_MODE); - if (dof->dof_type != DFT_NODE) - attr->la_valid &= ~LA_RDEV; - if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime))) - attr->la_valid &= ~LA_ATIME; - if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime))) - attr->la_valid &= ~LA_CTIME; - if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime))) - attr->la_valid &= ~LA_MTIME; + if (dof->dof_type != DFT_NODE) + attr->la_valid &= ~LA_RDEV; + if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime))) + attr->la_valid &= ~LA_ATIME; + if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime))) + attr->la_valid &= ~LA_CTIME; + if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime))) + attr->la_valid &= ~LA_MTIME; result = osd_quota_transfer(inode, attr); if (result) @@ -3050,8 +3303,8 @@ static int __osd_create(struct osd_thread_info *info, struct osd_object *obj, struct lu_attr *attr, struct dt_allocation_hint *hint, struct dt_object_format *dof, struct thandle *th) { - int result; - __u32 umask; + int result; + __u32 umask; osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE); @@ -3064,8 +3317,10 @@ static int __osd_create(struct osd_thread_info *info, struct osd_object *obj, if (likely(obj->oo_inode != NULL)) { LASSERT(obj->oo_inode->i_state & I_NEW); - /* Unlock the inode before attr initialization to avoid - * unnecessary dqget operations. LU-6378 */ + /* + * Unlock the inode before attr initialization to avoid + * unnecessary dqget operations. LU-6378 + */ unlock_new_inode(obj->oo_inode); } @@ -3094,7 +3349,7 @@ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj, struct osd_inode_id *id = &info->oti_id; struct osd_device *osd = osd_obj2dev(obj); struct osd_thandle *oh; - int rc; + int rc; LASSERT(obj->oo_inode != NULL); @@ -3113,7 +3368,7 @@ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj, int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd, u64 seq, struct lu_seq_range *range) { - struct seq_server_site *ss = osd_seq_site(osd); + struct seq_server_site *ss = osd_seq_site(osd); if (fid_seq_is_idif(seq)) { fld_range_set_ost(range); @@ -3124,8 +3379,10 @@ int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd, if (!fid_seq_in_fldb(seq)) { fld_range_set_mdt(range); if (ss != NULL) - /* FIXME: If ss is NULL, it suppose not get lsr_index - * at all */ + /* + * FIXME: If ss is NULL, it suppose not get lsr_index + * at all + */ range->lsr_index = ss->ss_node_id; return 0; } @@ -3144,6 +3401,7 @@ static int osd_declare_create(const struct lu_env *env, struct dt_object *dt, { struct osd_thandle *oh; int rc; + ENTRY; LASSERT(handle != NULL); @@ -3151,18 +3409,28 @@ static int osd_declare_create(const struct lu_env *env, struct dt_object *dt, oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - /* EA object consumes more credits than regular object: osd_mk_index + /* + * EA object consumes more credits than regular object: osd_mk_index * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and * leaf_node, could involves the block, block bitmap, groups, GDT - * change for each block, so add 4 * 2 credits in that case. */ + * change for each block, so add 4 * 2 credits in that case. + */ osd_trans_declare_op(env, oh, OSD_OT_CREATE, osd_dto_credits_noquota[DTO_OBJECT_CREATE] + (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0); - /* Reuse idle OI block may cause additional one OI block - * to be changed. */ + /* + * Reuse idle OI block may cause additional one OI block + * to be changed. + */ osd_trans_declare_op(env, oh, OSD_OT_INSERT, osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1); + /* will help to find FID->ino mapping at dt_insert() */ + rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)), + osd_dt_obj(dt)); + if (rc != 0) + RETURN(rc); + if (!attr) RETURN(0); @@ -3172,10 +3440,6 @@ static int osd_declare_create(const struct lu_env *env, struct dt_object *dt, if (rc != 0) RETURN(rc); - /* will help to find FID->ino mapping at dt_insert() */ - rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)), - osd_dt_obj(dt)); - RETURN(rc); } @@ -3187,10 +3451,11 @@ static int osd_declare_create(const struct lu_env *env, struct dt_object *dt, static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt, struct thandle *th) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; struct osd_thandle *oh; - int rc; + int rc; + ENTRY; if (inode == NULL) @@ -3206,8 +3471,10 @@ static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt, if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE]; - /* Recycle idle OI leaf may cause additional three OI blocks - * to be changed. */ + /* + * Recycle idle OI leaf may cause additional three OI blocks + * to be changed. + */ if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2)) osd_trans_declare_op(env, oh, OSD_OT_DELETE, osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3); @@ -3224,8 +3491,10 @@ static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt, if (rc) RETURN(rc); - /* will help to find FID->ino when this object is being - * added to PENDING/ */ + /* + * will help to find FID->ino when this object is being + * added to PENDING + */ rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj); RETURN(rc); @@ -3240,6 +3509,7 @@ static int osd_destroy(const struct lu_env *env, struct dt_object *dt, struct osd_device *osd = osd_obj2dev(obj); struct osd_thandle *oh; int result; + ENTRY; oh = container_of0(th, struct osd_thandle, ot_super); @@ -3299,9 +3569,10 @@ static int osd_destroy(const struct lu_env *env, struct dt_object *dt, int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, const struct lu_fid *fid, __u32 compat, __u32 incompat) { - struct lustre_ost_attrs *loa = &info->oti_ost_attrs; - struct lustre_mdt_attrs *lma = &loa->loa_lma; + struct lustre_ost_attrs *loa = &info->oti_ost_attrs; + struct lustre_mdt_attrs *lma = &loa->loa_lma; int rc; + ENTRY; if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA)) @@ -3313,10 +3584,12 @@ int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, lustre_loa_init(loa, fid, compat, incompat); lustre_loa_swab(loa, false); - /* For the OST device with 256 bytes inode size by default, + /* + * For the OST device with 256 bytes inode size by default, * the PFID EA will be stored together with LMA EA to avoid * performance trouble. Otherwise the PFID EA can be stored - * independently. LU-8998 */ + * independently. LU-8998 + */ if ((compat & LMAC_FID_ON_OST) && LDISKFS_INODE_SIZE(inode->i_sb) <= 256) rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, @@ -3324,8 +3597,10 @@ int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, else rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma), XATTR_CREATE); - /* LMA may already exist, but we need to check that all the - * desired compat/incompat flags have been added. */ + /* + * LMA may already exist, but we need to check that all the + * desired compat/incompat flags have been added. + */ if (unlikely(rc == -EEXIST)) { rc = __osd_xattr_get(inode, &info->oti_obj_dentry, XATTR_NAME_LMA, (void *)loa, sizeof(*loa)); @@ -3393,7 +3668,8 @@ static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj, struct osd_inode_id *id) { struct osd_thread_info *info = osd_oti_get(env); - struct inode *inode; + struct inode *inode; + ENTRY; osd_id_gen(id, ino, OSD_OII_NOGEN); @@ -3425,7 +3701,8 @@ static int osd_add_dot_dotdot_internal(struct osd_thread_info *info, rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir, dir, dot_ldp, dot_dot_ldp); - /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then + /* + * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then * the subseqent ref_add() will increase the dir->i_nlink * as 3. That is incorrect for new created directory. * @@ -3437,7 +3714,8 @@ static int osd_add_dot_dotdot_internal(struct osd_thread_info *info, * becuase if the nlink attr is not 2, then there will be * ref_add() called following the dt_insert(..), such call * will make both the in-RAM and on-disk dir-inode's nlink - * attr to be set as 2. LU-7447 */ + * attr to be set as 2. LU-7447 + */ set_nlink(dir, saved_nlink); return rc; } @@ -3452,10 +3730,11 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, __u32 type, struct thandle *th) { - struct osd_thread_info *info = osd_oti_get(env); - struct inode *local; - struct osd_thandle *oh; - int rc; + struct osd_thread_info *info = osd_oti_get(env); + struct inode *local; + struct osd_thandle *oh; + int rc; + ENTRY; LASSERT(th); @@ -3469,8 +3748,10 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, RETURN(local); } - /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set - * correct gid on remote file, not agent here */ + /* + * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set + * correct gid on remote file, not agent here + */ local->i_gid = current_fsgid(); ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB); unlock_new_inode(local); @@ -3479,7 +3760,7 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, #ifdef HAVE_PROJECT_QUOTA if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL && i_projid_read(pobj->oo_inode) != 0) { - rc = ldiskfs_transfer_project(local, 0); + rc = osd_transfer_project(local, 0); if (rc) { CERROR("%s: quota transfer failed: rc = %d. Is project " "quota enforcement enabled on the ldiskfs " @@ -3525,7 +3806,7 @@ static int osd_schedule_agent_inode_removal(const struct lu_env *env, struct osd_thandle *oh, __u32 ino) { - struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev); + struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev); struct osd_obj_orphan *oor; OBD_ALLOC_PTR(oor); @@ -3606,10 +3887,10 @@ static int osd_create(const struct lu_env *env, struct dt_object *dt, struct lu_attr *attr, struct dt_allocation_hint *hint, struct dt_object_format *dof, struct thandle *th) { - const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thread_info *info = osd_oti_get(env); - int result, on_ost = 0; + const struct lu_fid *fid = lu_object_fid(&dt->do_lu); + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thread_info *info = osd_oti_get(env); + int result, on_ost = 0; ENTRY; @@ -3622,8 +3903,10 @@ static int osd_create(const struct lu_env *env, struct dt_object *dt, LASSERT(th != NULL); if (unlikely(fid_is_acct(fid))) - /* Quota files can't be created from the kernel any more, - * 'tune2fs -O quota' will take care of creating them */ + /* + * Quota files can't be created from the kernel any more, + * 'tune2fs -O quota' will take care of creating them + */ RETURN(-EPERM); result = __osd_create(info, obj, attr, hint, dof, th); @@ -3649,14 +3932,17 @@ static int osd_create(const struct lu_env *env, struct dt_object *dt, obj->oo_dt.do_body_ops = &osd_body_ops; } - if (result == 0) + if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY)) result = __osd_oi_insert(env, obj, fid, th); - /* a small optimization - dt_insert() isn't usually applied + /* + * a small optimization - dt_insert() isn't usually applied * to OST objects, so we don't need to cache OI mapping for - * OST objects */ + * OST objects + */ if (result == 0 && on_ost == 0) { struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + result = osd_idc_find_and_init(env, osd, obj); LASSERT(result == 0); } @@ -3681,6 +3967,8 @@ static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt, osd_trans_declare_op(env, oh, OSD_OT_REF_ADD, osd_dto_credits_noquota[DTO_ATTR_SET_BASE]); + osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt)); + return 0; } @@ -3690,10 +3978,10 @@ static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt, static int osd_ref_add(const struct lu_env *env, struct dt_object *dt, struct thandle *th) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; struct osd_thandle *oh; - int rc = 0; + int rc = 0; if (!dt_object_exists(dt) || obj->oo_destroyed) return -ENOENT; @@ -3766,10 +4054,10 @@ static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt, static int osd_ref_del(const struct lu_env *env, struct dt_object *dt, struct thandle *th) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - struct osd_thandle *oh; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + struct osd_thandle *oh; if (!dt_object_exists(dt)) return -ENOENT; @@ -3785,9 +4073,11 @@ static int osd_ref_del(const struct lu_env *env, struct dt_object *dt, osd_trans_exec_op(env, th, OSD_OT_REF_DEL); spin_lock(&obj->oo_guard); - /* That can be result of upgrade from old Lustre version and + /* + * That can be result of upgrade from old Lustre version and * applied only to local files. Just skip this ref_del call. - * ext4_unlink() only treats this as a warning, don't LASSERT here.*/ + * ext4_unlink() only treats this as a warning, don't LASSERT here. + */ if (inode->i_nlink == 0) { CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ? D_ERROR : D_INODE, "%s: nlink == 0 on "DFID @@ -3821,8 +4111,8 @@ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, struct inode *inode = obj->oo_inode; struct osd_thread_info *info = osd_oti_get(env); struct dentry *dentry = &info->oti_obj_dentry; - bool cache_xattr = false; - int rc; + bool cache_xattr = false; + int rc; LASSERT(buf); @@ -3830,8 +4120,10 @@ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, if (strcmp(name, XATTR_NAME_VERSION) == 0) { dt_obj_version_t *ver = buf->lb_buf; - /* for version we are just using xattr API but change inode - * field instead */ + /* + * for version we are just using xattr API but change inode + * field instead + */ if (buf->lb_len == 0) return sizeof(dt_obj_version_t); @@ -3851,7 +4143,9 @@ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, LASSERT(!dt_object_remote(dt)); LASSERT(inode->i_op != NULL); +#ifdef HAVE_IOP_XATTR LASSERT(inode->i_op->getxattr != NULL); +#endif if (strcmp(name, XATTR_NAME_LOV) == 0 || strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) @@ -3914,9 +4208,9 @@ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, } static int osd_declare_xattr_set(const struct lu_env *env, - struct dt_object *dt, - const struct lu_buf *buf, const char *name, - int fl, struct thandle *handle) + struct dt_object *dt, + const struct lu_buf *buf, const char *name, + int fl, struct thandle *handle) { struct osd_thandle *oh; int credits = 0; @@ -3928,9 +4222,11 @@ static int osd_declare_xattr_set(const struct lu_env *env, LASSERT(oh->ot_handle == NULL); if (strcmp(name, XATTR_NAME_LMA) == 0) { - /* For non-upgrading case, the LMA is set first and + /* + * For non-upgrading case, the LMA is set first and * usually fit inode. But for upgrade case, the LMA - * may be in another separated EA block. */ + * may be in another separated EA block. + */ if (dt_object_exists(dt)) { if (fl == LU_XATTR_REPLACE) credits = 1; @@ -3947,12 +4243,14 @@ static int osd_declare_xattr_set(const struct lu_env *env, else goto upgrade; } else { - /* If some name entry resides on remote MDT, then will create + /* + * If some name entry resides on remote MDT, then will create * agent entry under remote parent. On the other hand, if the * remote entry will be removed, then related agent entry may * need to be removed from the remote parent. So there may be * kinds of cases, let's declare enough credits. The credits - * for create agent entry is enough for remove case. */ + * for create agent entry is enough for remove case. + */ if (strcmp(name, XATTR_NAME_LINK) == 0) { credits += osd_dto_credits_noquota[DTO_INDEX_INSERT]; if (dt_object_exists(dt)) @@ -3966,8 +4264,10 @@ upgrade: ssize_t buflen; if (buf->lb_buf == NULL && dt_object_exists(dt)) { - /* learn xattr size from osd_xattr_get if - attribute has not been read yet */ + /* + * learn xattr size from osd_xattr_get if + * attribute has not been read yet + */ buflen = __osd_xattr_get( osd_dt_obj(dt)->oo_inode, &osd_oti_get(env)->oti_obj_dentry, @@ -4010,6 +4310,7 @@ static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj, struct ost_layout *ol = &ff->ff_layout; int flags = XATTR_REPLACE; int rc; + ENTRY; if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid)) @@ -4046,15 +4347,17 @@ static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj, if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) RETURN(fl); - /* Old client does not send stripe information, - * then store the PFID EA on disk separatedly. */ + /* + * Old client does not send stripe information, + * then store the PFID EA on disk separatedly. + */ if (unlikely(buf->lb_len == sizeof(struct lu_fid) || ol->ol_stripe_size == 0)) RETURN(fl); /* Remove old PFID EA entry firstly. */ ll_vfs_dq_init(inode); - rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID); + rc = osd_removexattr(dentry, inode, XATTR_NAME_FID); if (rc == -ENODATA) { if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE)) RETURN(rc); @@ -4131,6 +4434,7 @@ static int osd_xattr_handle_linkea(const struct lu_env *env, struct osd_thandle *oh; int rc; bool remote = false; + ENTRY; oh = container_of0(handle, struct osd_thandle, ot_super); @@ -4182,6 +4486,7 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, int fs_flags = 0; int len; int rc; + ENTRY; LASSERT(handle); @@ -4191,16 +4496,20 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, if (strcmp(name, XATTR_NAME_VERSION) == 0) { dt_obj_version_t *version = buf->lb_buf; - /* for version we are just using xattr API but change inode - * field instead */ + /* + * for version we are just using xattr API but change inode + * field instead + */ LASSERT(buf->lb_len == sizeof(dt_obj_version_t)); CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n", *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino); LDISKFS_I(inode)->i_fs_version = *version; - /* Version is set after all inode operations are finished, - * so we should mark it dirty here */ + /* + * Version is set after all inode operations are finished, + * so we should mark it dirty here + */ ll_dirty_inode(inode, I_DIRTY_DATASYNC); RETURN(0); @@ -4212,10 +4521,12 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, len = buf->lb_len; osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); - /* For the OST device with 256 bytes inode size by default, + /* + * For the OST device with 256 bytes inode size by default, * the PFID EA will be stored together with LMA EA to avoid * performance trouble. Otherwise the PFID EA can be stored - * independently. LU-8998 */ + * independently. LU-8998 + */ if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost && (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) { LASSERT(buf->lb_buf); @@ -4268,10 +4579,10 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt, const struct lu_buf *buf) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_thread_info *info = osd_oti_get(env); - struct dentry *dentry = &info->oti_obj_dentry; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct dentry *dentry = &info->oti_obj_dentry; if (!dt_object_exists(dt)) return -ENOENT; @@ -4315,19 +4626,21 @@ static int osd_declare_xattr_del(const struct lu_env *env, static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt, const char *name, struct thandle *handle) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_thread_info *info = osd_oti_get(env); - struct dentry *dentry = &info->oti_obj_dentry; - int rc; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct dentry *dentry = &info->oti_obj_dentry; + int rc; if (!dt_object_exists(dt)) return -ENOENT; LASSERT(!dt_object_remote(dt)); LASSERT(inode->i_op != NULL); - LASSERT(inode->i_op->removexattr != NULL); LASSERT(handle != NULL); +#ifdef HAVE_IOP_XATTR + LASSERT(inode->i_op->removexattr != NULL); +#endif osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); @@ -4350,7 +4663,7 @@ static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt, ll_vfs_dq_init(inode); dentry->d_inode = inode; dentry->d_sb = inode->i_sb; - rc = inode->i_op->removexattr(dentry, name); + rc = osd_removexattr(dentry, inode, name); } osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET); @@ -4366,12 +4679,12 @@ static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt, static int osd_object_sync(const struct lu_env *env, struct dt_object *dt, __u64 start, __u64 end) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_thread_info *info = osd_oti_get(env); - struct dentry *dentry = &info->oti_obj_dentry; - struct file *file = &info->oti_file; - int rc; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct dentry *dentry = &info->oti_obj_dentry; + struct file *file = &info->oti_file; + int rc; ENTRY; @@ -4397,53 +4710,51 @@ static int osd_invalidate(const struct lu_env *env, struct dt_object *dt) */ static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o, - const struct dt_index_features *feat) -{ - struct iam_descr *descr; - - if (osd_object_is_root(o)) - return feat == &dt_directory_features; - - LASSERT(o->oo_dir != NULL); - - descr = o->oo_dir->od_container.ic_descr; - if (feat == &dt_directory_features) { - if (descr->id_rec_size == sizeof(struct osd_fid_pack)) - return 1; - else - return 0; - } else { - return - feat->dif_keysize_min <= descr->id_key_size && - descr->id_key_size <= feat->dif_keysize_max && - feat->dif_recsize_min <= descr->id_rec_size && - descr->id_rec_size <= feat->dif_recsize_max && - !(feat->dif_flags & (DT_IND_VARKEY | - DT_IND_VARREC | DT_IND_NONUNQ)) && - ergo(feat->dif_flags & DT_IND_UPDATE, - 1 /* XXX check that object (and file system) is - * writable */); - } + const struct dt_index_features *feat) +{ + struct iam_descr *descr; + + if (osd_object_is_root(o)) + return feat == &dt_directory_features; + + LASSERT(o->oo_dir != NULL); + + descr = o->oo_dir->od_container.ic_descr; + if (feat == &dt_directory_features) { + if (descr->id_rec_size == sizeof(struct osd_fid_pack)) + return 1; + else + return 0; + } else { + return feat->dif_keysize_min <= descr->id_key_size && + descr->id_key_size <= feat->dif_keysize_max && + feat->dif_recsize_min <= descr->id_rec_size && + descr->id_rec_size <= feat->dif_recsize_max && + !(feat->dif_flags & (DT_IND_VARKEY | + DT_IND_VARREC | DT_IND_NONUNQ)) && + ergo(feat->dif_flags & DT_IND_UPDATE, + 1 /* XXX check that object (and fs) is writable */); + } } static int osd_iam_container_init(const struct lu_env *env, - struct osd_object *obj, - struct osd_directory *dir) + struct osd_object *obj, + struct osd_directory *dir) { - struct iam_container *bag = &dir->od_container; - int result; + struct iam_container *bag = &dir->od_container; + int result; - result = iam_container_init(bag, &dir->od_descr, obj->oo_inode); - if (result != 0) - return result; + result = iam_container_init(bag, &dir->od_descr, obj->oo_inode); + if (result != 0) + return result; - result = iam_container_setup(bag); - if (result == 0) - obj->oo_dt.do_index_ops = &osd_index_iam_ops; - else - iam_container_fini(bag); + result = iam_container_setup(bag); + if (result == 0) + obj->oo_dt.do_index_ops = &osd_index_iam_ops; + else + iam_container_fini(bag); - return result; + return result; } @@ -4451,23 +4762,23 @@ static int osd_iam_container_init(const struct lu_env *env, * Concurrency: no external locking is necessary. */ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, - const struct dt_index_features *feat) + const struct dt_index_features *feat) { - int result; - int skip_iam = 0; - struct osd_object *obj = osd_dt_obj(dt); + int result; + int skip_iam = 0; + struct osd_object *obj = osd_dt_obj(dt); - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); - if (osd_object_is_root(obj)) { - dt->do_index_ops = &osd_index_ea_ops; - result = 0; + if (osd_object_is_root(obj)) { + dt->do_index_ops = &osd_index_ea_ops; + result = 0; } else if (feat == &dt_directory_features) { - dt->do_index_ops = &osd_index_ea_ops; + dt->do_index_ops = &osd_index_ea_ops; if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode)) - result = 0; - else - result = -ENOTDIR; + result = 0; + else + result = -ENOTDIR; skip_iam = 1; } else if (unlikely(feat == &dt_otable_features)) { dt->do_index_ops = &osd_otable_ops; @@ -4476,11 +4787,13 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, dt->do_index_ops = &osd_acct_index_ops; result = 0; skip_iam = 1; - } else if (!osd_has_index(obj)) { - struct osd_directory *dir; + } else if (!osd_has_index(obj)) { + struct osd_directory *dir; + struct osd_device *osd = osd_obj2dev(obj); + const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - OBD_ALLOC_PTR(dir); - if (dir != NULL) { + OBD_ALLOC_PTR(dir); + if (dir) { spin_lock(&obj->oo_guard); if (obj->oo_dir == NULL) @@ -4499,26 +4812,50 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, /* * recheck under lock. */ - if (!osd_has_index(obj)) - result = osd_iam_container_init(env, obj, - obj->oo_dir); - else + + if (osd_has_index(obj)) { + result = 0; + goto unlock; + } + + result = osd_iam_container_init(env, obj, obj->oo_dir); + if (result || feat == &dt_lfsck_namespace_features || + feat == &dt_lfsck_layout_orphan_features || + feat == &dt_lfsck_layout_dangling_features) + goto unlock; + + result = osd_index_register(osd, fid, + feat->dif_keysize_max, + feat->dif_recsize_max); + if (result < 0) + CWARN("%s: failed to register index " + DFID": rc = %d\n", + osd_name(osd), PFID(fid), result); + else if (result > 0) result = 0; + else + CDEBUG(D_LFSCK, "%s: index object "DFID + " (%d/%d) registered\n", + osd_name(osd), PFID(fid), + (int)feat->dif_keysize_max, + (int)feat->dif_recsize_max); + +unlock: up_write(&obj->oo_ext_idx_sem); - } else { - result = -ENOMEM; - } - } else { - result = 0; - } + } else { + result = -ENOMEM; + } + } else { + result = 0; + } if (result == 0 && skip_iam == 0) { - if (!osd_iam_index_probe(env, obj, feat)) - result = -ENOTDIR; - } - LINVRNT(osd_invariant(obj)); + if (!osd_iam_index_probe(env, obj, feat)) + result = -ENOTDIR; + } + LINVRNT(osd_invariant(obj)); - return result; + return result; } static int osd_otable_it_attr_get(const struct lu_env *env, @@ -4564,14 +4901,14 @@ static const struct dt_object_operations osd_obj_otable_it_ops = { }; static int osd_index_declare_iam_delete(const struct lu_env *env, - struct dt_object *dt, - const struct dt_key *key, - struct thandle *handle) + struct dt_object *dt, + const struct dt_key *key, + struct thandle *handle) { - struct osd_thandle *oh; + struct osd_thandle *oh; - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle == NULL); + oh = container_of0(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle == NULL); /* Recycle may cause additional three blocks to be changed. */ osd_trans_declare_op(env, oh, OSD_OT_DELETE, @@ -4596,11 +4933,12 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, struct thandle *handle) { struct osd_thread_info *oti = osd_oti_get(env); - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thandle *oh; - struct iam_path_descr *ipd; - struct iam_container *bag = &obj->oo_dir->od_container; - int rc; + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thandle *oh; + struct iam_path_descr *ipd; + struct iam_container *bag = &obj->oo_dir->od_container; + int rc; + ENTRY; if (!dt_object_exists(dt)) @@ -4613,13 +4951,13 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, osd_trans_exec_op(env, handle, OSD_OT_DELETE); - ipd = osd_idx_ipd_get(env, bag); - if (unlikely(ipd == NULL)) - RETURN(-ENOMEM); + ipd = osd_idx_ipd_get(env, bag); + if (unlikely(ipd == NULL)) + RETURN(-ENOMEM); - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle != NULL); - LASSERT(oh->ot_handle->h_transaction != NULL); + oh = container_of0(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); + LASSERT(oh->ot_handle->h_transaction != NULL); if (fid_is_quota(lu_object_fid(&dt->do_lu))) { /* swab quota uid/gid provided by caller */ @@ -4627,11 +4965,11 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, key = (const struct dt_key *)&oti->oti_quota_id; } - rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd); - osd_ipd_put(env, bag, ipd); - LINVRNT(osd_invariant(obj)); + rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd); + osd_ipd_put(env, bag, ipd); + LINVRNT(osd_invariant(obj)); osd_trans_exec_check(env, handle, OSD_OT_DELETE); - RETURN(rc); + RETURN(rc); } static int osd_index_declare_ea_delete(const struct lu_env *env, @@ -4640,8 +4978,9 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, struct thandle *handle) { struct osd_thandle *oh; - struct inode *inode; - int rc, credits; + struct inode *inode; + int rc, credits; + ENTRY; LASSERT(!dt_object_remote(dt)); @@ -4664,33 +5003,36 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, } static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de, - struct dt_rec *fid) + struct dt_rec *fid) { - struct osd_fid_pack *rec; - int rc = -ENODATA; + struct osd_fid_pack *rec; + int rc = -ENODATA; - if (de->file_type & LDISKFS_DIRENT_LUFID) { - rec = (struct osd_fid_pack *) (de->name + de->name_len + 1); - rc = osd_fid_unpack((struct lu_fid *)fid, rec); + if (de->file_type & LDISKFS_DIRENT_LUFID) { + rec = (struct osd_fid_pack *)(de->name + de->name_len + 1); + rc = osd_fid_unpack((struct lu_fid *)fid, rec); if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid))) rc = -EINVAL; - } + } return rc; } static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, const struct lu_fid *fid) { - struct seq_server_site *ss = osd_seq_site(osd); + struct seq_server_site *ss = osd_seq_site(osd); + ENTRY; /* FID seqs not in FLDB, must be local seq */ if (unlikely(!fid_seq_in_fldb(fid_seq(fid)))) RETURN(0); - /* If FLD is not being initialized yet, it only happens during the + /* + * If FLD is not being initialized yet, it only happens during the * initialization, likely during mgs initialization, and we assume - * this is local FID. */ + * this is local FID. + */ if (ss == NULL || ss->ss_server_fld == NULL) RETURN(0); @@ -4701,6 +5043,36 @@ static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, RETURN(1); } +static void osd_take_care_of_agent(const struct lu_env *env, + struct osd_device *osd, + struct osd_thandle *oh, + struct ldiskfs_dir_entry_2 *de) +{ + struct lu_fid *fid = &osd_oti_get(env)->oti_fid; + struct osd_idmap_cache *idc; + int rc, schedule = 0; + + LASSERT(de != NULL); + + rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid); + if (likely(rc == 0)) { + idc = osd_idc_find_or_init(env, osd, fid); + if (IS_ERR(idc) || idc->oic_remote) + schedule = 1; + } else if (rc == -ENODATA) { + /* + * can't get FID, postpone to the end of the + * transaction when iget() is safe + */ + schedule = 1; + } else { + CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc); + } + if (schedule) + osd_schedule_agent_inode_removal(env, oh, + le32_to_cpu(de->inode)); +} + /** * Index delete function for interoperability mode (b11826). * It will remove the directory entry added by osd_index_ea_insert(). @@ -4714,16 +5086,16 @@ static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, const struct dt_key *key, struct thandle *handle) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *dir = obj->oo_inode; - struct dentry *dentry; - struct osd_thandle *oh; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *dir = obj->oo_inode; + struct dentry *dentry; + struct osd_thandle *oh; struct ldiskfs_dir_entry_2 *de = NULL; - struct buffer_head *bh; - struct htree_lock *hlock = NULL; - struct lu_fid *fid = &osd_oti_get(env)->oti_fid; - struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - int rc; + struct buffer_head *bh; + struct htree_lock *hlock = NULL; + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + int rc; + ENTRY; if (!dt_object_exists(dt)) @@ -4735,25 +5107,26 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, osd_trans_exec_op(env, handle, OSD_OT_DELETE); - oh = container_of(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle != NULL); - LASSERT(oh->ot_handle->h_transaction != NULL); + oh = container_of(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); + LASSERT(oh->ot_handle->h_transaction != NULL); ll_vfs_dq_init(dir); - dentry = osd_child_dentry_get(env, obj, - (char *)key, strlen((char *)key)); - - if (obj->oo_hl_head != NULL) { - hlock = osd_oti_get(env)->oti_hlock; - ldiskfs_htree_lock(hlock, obj->oo_hl_head, - dir, LDISKFS_HLOCK_DEL); - } else { + dentry = osd_child_dentry_get(env, obj, + (char *)key, strlen((char *)key)); + + if (obj->oo_hl_head != NULL) { + hlock = osd_oti_get(env)->oti_hlock; + ldiskfs_htree_lock(hlock, obj->oo_hl_head, + dir, LDISKFS_HLOCK_DEL); + } else { down_write(&obj->oo_ext_idx_sem); - } + } - bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); + bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); if (!IS_ERR(bh)) { - /* If this is not the ".." entry, it might be a remote DNE + /* + * If this is not the ".." entry, it might be a remote DNE * entry and we need to check if the FID is for a remote * MDT. If the FID is not in the directory entry (e.g. * upgraded 1.8 filesystem without dirdata enabled) then @@ -4766,37 +5139,25 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, * the entry first might leak the agent inode afterward. The * reverse would need filesystem abort in case of error deleting * the entry after the agent had been removed, or leave a - * dangling entry pointing at a random inode. */ - if (strcmp((char *)key, dotdot) != 0) { - LASSERT(de != NULL); - rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid); - if (rc == -ENODATA) { - /* can't get FID, postpone to the end of the - * transaction when iget() is safe */ - osd_schedule_agent_inode_removal(env, oh, - le32_to_cpu(de->inode)); - } else if (rc == 0 && - unlikely(osd_remote_fid(env, osd, fid))) { - osd_schedule_agent_inode_removal(env, oh, - le32_to_cpu(de->inode)); - } - } + * dangling entry pointing at a random inode. + */ + if (strcmp((char *)key, dotdot) != 0) + osd_take_care_of_agent(env, osd, oh, de); rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh); brelse(bh); } else { rc = PTR_ERR(bh); } - if (hlock != NULL) - ldiskfs_htree_unlock(hlock); - else + if (hlock != NULL) + ldiskfs_htree_unlock(hlock); + else up_write(&obj->oo_ext_idx_sem); GOTO(out, rc); - out: - LASSERT(osd_invariant(obj)); + LASSERT(osd_invariant(obj)); osd_trans_exec_check(env, handle, OSD_OT_DELETE); - RETURN(rc); + RETURN(rc); } /** @@ -4813,13 +5174,14 @@ out: static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt, struct dt_rec *rec, const struct dt_key *key) { - struct osd_object *obj = osd_dt_obj(dt); - struct iam_path_descr *ipd; - struct iam_container *bag = &obj->oo_dir->od_container; + struct osd_object *obj = osd_dt_obj(dt); + struct iam_path_descr *ipd; + struct iam_container *bag = &obj->oo_dir->od_container; struct osd_thread_info *oti = osd_oti_get(env); - struct iam_iterator *it = &oti->oti_idx_it; - struct iam_rec *iam_rec; - int rc; + struct iam_iterator *it = &oti->oti_idx_it; + struct iam_rec *iam_rec; + int rc; + ENTRY; if (!dt_object_exists(dt)) @@ -4829,12 +5191,12 @@ static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt, LASSERT(!dt_object_remote(dt)); LASSERT(bag->ic_object == obj->oo_inode); - ipd = osd_idx_ipd_get(env, bag); - if (IS_ERR(ipd)) - RETURN(-ENOMEM); + ipd = osd_idx_ipd_get(env, bag); + if (IS_ERR(ipd)) + RETURN(-ENOMEM); - /* got ipd now we can start iterator. */ - iam_it_init(it, bag, 0, ipd); + /* got ipd now we can start iterator. */ + iam_it_init(it, bag, 0, ipd); if (fid_is_quota(lu_object_fid(&dt->do_lu))) { /* swab quota uid/gid provided by caller */ @@ -4842,43 +5204,43 @@ static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt, key = (const struct dt_key *)&oti->oti_quota_id; } - rc = iam_it_get(it, (struct iam_key *)key); - if (rc >= 0) { - if (S_ISDIR(obj->oo_inode->i_mode)) - iam_rec = (struct iam_rec *)oti->oti_ldp; - else - iam_rec = (struct iam_rec *) rec; + rc = iam_it_get(it, (struct iam_key *)key); + if (rc >= 0) { + if (S_ISDIR(obj->oo_inode->i_mode)) + iam_rec = (struct iam_rec *)oti->oti_ldp; + else + iam_rec = (struct iam_rec *)rec; - iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec); + iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec); - if (S_ISDIR(obj->oo_inode->i_mode)) - osd_fid_unpack((struct lu_fid *) rec, - (struct osd_fid_pack *)iam_rec); + if (S_ISDIR(obj->oo_inode->i_mode)) + osd_fid_unpack((struct lu_fid *)rec, + (struct osd_fid_pack *)iam_rec); else if (fid_is_quota(lu_object_fid(&dt->do_lu))) osd_quota_unpack(obj, rec); - } + } - iam_it_put(it); - iam_it_fini(it); - osd_ipd_put(env, bag, ipd); + iam_it_put(it); + iam_it_fini(it); + osd_ipd_put(env, bag, ipd); - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); - RETURN(rc); + RETURN(rc); } static int osd_index_declare_iam_insert(const struct lu_env *env, - struct dt_object *dt, - const struct dt_rec *rec, - const struct dt_key *key, - struct thandle *handle) + struct dt_object *dt, + const struct dt_rec *rec, + const struct dt_key *key, + struct thandle *handle) { - struct osd_thandle *oh; + struct osd_thandle *oh; - LASSERT(handle != NULL); + LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle == NULL); + oh = container_of0(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle == NULL); osd_trans_declare_op(env, oh, OSD_OT_INSERT, osd_dto_credits_noquota[DTO_INDEX_INSERT]); @@ -4902,13 +5264,14 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, const struct dt_key *key, struct thandle *th, int ignore_quota) { - struct osd_object *obj = osd_dt_obj(dt); + struct osd_object *obj = osd_dt_obj(dt); struct iam_path_descr *ipd; - struct osd_thandle *oh; - struct iam_container *bag; + struct osd_thandle *oh; + struct iam_container *bag; struct osd_thread_info *oti = osd_oti_get(env); - struct iam_rec *iam_rec; - int rc; + struct iam_rec *iam_rec; + int rc; + ENTRY; if (!dt_object_exists(dt)) @@ -4923,16 +5286,17 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, osd_trans_exec_op(env, th, OSD_OT_INSERT); - ipd = osd_idx_ipd_get(env, bag); - if (unlikely(ipd == NULL)) - RETURN(-ENOMEM); + ipd = osd_idx_ipd_get(env, bag); + if (unlikely(ipd == NULL)) + RETURN(-ENOMEM); - oh = container_of0(th, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle != NULL); - LASSERT(oh->ot_handle->h_transaction != NULL); + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); + LASSERT(oh->ot_handle->h_transaction != NULL); if (S_ISDIR(obj->oo_inode->i_mode)) { iam_rec = (struct iam_rec *)oti->oti_ldp; - osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid); + osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, + &oti->oti_fid); } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) { /* pack quota uid/gid */ oti->oti_quota_id = cpu_to_le64(*((__u64 *)key)); @@ -4944,12 +5308,12 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, iam_rec = (struct iam_rec *)rec; } - rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key, - iam_rec, ipd); - osd_ipd_put(env, bag, ipd); - LINVRNT(osd_invariant(obj)); + rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key, + iam_rec, ipd); + osd_ipd_put(env, bag, ipd); + LINVRNT(osd_invariant(obj)); osd_trans_exec_check(env, th, OSD_OT_INSERT); - RETURN(rc); + RETURN(rc); } /** @@ -4965,14 +5329,14 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, const char *name, const struct lu_fid *fid, struct htree_lock *hlock, struct thandle *th) { - struct ldiskfs_dentry_param *ldp; - struct dentry *child; - struct osd_thandle *oth; - int rc; + struct ldiskfs_dentry_param *ldp; + struct dentry *child; + struct osd_thandle *oth; + int rc; - oth = container_of(th, struct osd_thandle, ot_super); - LASSERT(oth->ot_handle != NULL); - LASSERT(oth->ot_handle->h_transaction != NULL); + oth = container_of(th, struct osd_thandle, ot_super); + LASSERT(oth->ot_handle != NULL); + LASSERT(oth->ot_handle->h_transaction != NULL); LASSERT(pobj->oo_inode); ldp = (struct ldiskfs_dentry_param *)info->oti_ldp; @@ -5032,22 +5396,22 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, const struct lu_fid *dot_dot_fid, struct thandle *th) { - struct inode *inode = dir->oo_inode; - struct osd_thandle *oth; - int result = 0; + struct inode *inode = dir->oo_inode; + struct osd_thandle *oth; + int result = 0; - oth = container_of(th, struct osd_thandle, ot_super); - LASSERT(oth->ot_handle->h_transaction != NULL); - LASSERT(S_ISDIR(dir->oo_inode->i_mode)); + oth = container_of(th, struct osd_thandle, ot_super); + LASSERT(oth->ot_handle->h_transaction != NULL); + LASSERT(S_ISDIR(dir->oo_inode->i_mode)); - if (strcmp(name, dot) == 0) { - if (dir->oo_compat_dot_created) { - result = -EEXIST; - } else { + if (strcmp(name, dot) == 0) { + if (dir->oo_compat_dot_created) { + result = -EEXIST; + } else { LASSERT(inode->i_ino == parent_dir->i_ino); - dir->oo_compat_dot_created = 1; - result = 0; - } + dir->oo_compat_dot_created = 1; + result = 0; + } } else if (strcmp(name, dotdot) == 0) { if (!dir->oo_compat_dot_created) return -EINVAL; @@ -5086,31 +5450,31 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, struct inode *cinode, const char *name, const struct lu_fid *fid, struct thandle *th) { - struct osd_thread_info *info = osd_oti_get(env); - struct htree_lock *hlock; - int rc; + struct osd_thread_info *info = osd_oti_get(env); + struct htree_lock *hlock; + int rc; - hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL; + hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL; - if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' && - name[2] =='\0'))) { - if (hlock != NULL) { - ldiskfs_htree_lock(hlock, pobj->oo_hl_head, - pobj->oo_inode, 0); - } else { + if (name[0] == '.' && (name[1] == '\0' || + (name[1] == '.' && name[2] == '\0'))) { + if (hlock != NULL) { + ldiskfs_htree_lock(hlock, pobj->oo_hl_head, + pobj->oo_inode, 0); + } else { down_write(&pobj->oo_ext_idx_sem); } rc = osd_add_dot_dotdot(info, pobj, cinode, name, lu_object_fid(&pobj->oo_dt.do_lu), - fid, th); - } else { - if (hlock != NULL) { - ldiskfs_htree_lock(hlock, pobj->oo_hl_head, - pobj->oo_inode, LDISKFS_HLOCK_ADD); - } else { + fid, th); + } else { + if (hlock != NULL) { + ldiskfs_htree_lock(hlock, pobj->oo_hl_head, + pobj->oo_inode, LDISKFS_HLOCK_ADD); + } else { down_write(&pobj->oo_ext_idx_sem); - } + } if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) { struct lu_fid *tfid = &info->oti_fid; @@ -5123,13 +5487,13 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, rc = __osd_ea_add_rec(info, pobj, cinode, name, fid, hlock, th); } - } - if (hlock != NULL) - ldiskfs_htree_unlock(hlock); - else + } + if (hlock != NULL) + ldiskfs_htree_unlock(hlock); + else up_write(&pobj->oo_ext_idx_sem); - return rc; + return rc; } static int @@ -5143,6 +5507,7 @@ osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev, int once = 0; bool insert; int rc; + ENTRY; if (!fid_is_norm(fid) && !fid_is_igif(fid)) @@ -5153,9 +5518,8 @@ osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev, RETURN(0); if (dev->od_auto_scrub_interval == AS_NEVER || - cfs_time_before(ktime_get_real_seconds(), - scrub->os_file.sf_time_last_complete + - dev->od_auto_scrub_interval)) + ktime_get_real_seconds() < + scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval) RETURN(0); again: @@ -5179,8 +5543,10 @@ again: if (gen != OSD_OII_NOGEN) goto trigger; - /* The inode may has been reused by others, we do not know, - * leave it to be handled by subsequent osd_fid_lookup(). */ + /* + * The inode may has been reused by others, we do not know, + * leave it to be handled by subsequent osd_fid_lookup(). + */ GOTO(out, rc = 0); } else if (rc || osd_id_eq(id, &oti->oti_id)) { GOTO(out, rc); @@ -5201,10 +5567,12 @@ trigger: } rc = osd_oii_insert(dev, oic, insert); - /* There is race condition between osd_oi_lookup and OI scrub. + /* + * There is race condition between osd_oi_lookup and OI scrub. * The OI scrub finished just after osd_oi_lookup() failure. * Under such case, it is unnecessary to trigger OI scrub again, - * but try to call osd_oi_lookup() again. */ + * but try to call osd_oi_lookup() again. + */ if (unlikely(rc == -EAGAIN)) goto again; @@ -5297,12 +5665,13 @@ static int osd_get_pfid_from_linkea(const struct lu_env *env, struct osd_object *obj, struct lu_fid *fid) { - struct osd_thread_info *oti = osd_oti_get(env); - struct lu_buf *buf = &oti->oti_big_buf; - struct dentry *dentry = &oti->oti_obj_dentry; - struct inode *inode = obj->oo_inode; - struct linkea_data ldata = { NULL }; - int rc; + struct osd_thread_info *oti = osd_oti_get(env); + struct lu_buf *buf = &oti->oti_big_buf; + struct dentry *dentry = &oti->oti_obj_dentry; + struct inode *inode = obj->oo_inode; + struct linkea_data ldata = { NULL }; + int rc; + ENTRY; fid_zero(fid); @@ -5360,6 +5729,7 @@ static int osd_verify_ent_by_linkea(const struct lu_env *env, struct lu_name cname = { .ln_name = name, .ln_namelen = namelen }; int rc; + ENTRY; again: @@ -5401,14 +5771,15 @@ again: static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, struct dt_rec *rec, const struct dt_key *key) { - struct inode *dir = obj->oo_inode; - struct dentry *dentry; - struct ldiskfs_dir_entry_2 *de; - struct buffer_head *bh; - struct lu_fid *fid = (struct lu_fid *) rec; - struct htree_lock *hlock = NULL; - int ino; - int rc; + struct inode *dir = obj->oo_inode; + struct dentry *dentry; + struct ldiskfs_dir_entry_2 *de; + struct buffer_head *bh; + struct lu_fid *fid = (struct lu_fid *)rec; + struct htree_lock *hlock = NULL; + int ino; + int rc; + ENTRY; LASSERT(dir->i_op != NULL); @@ -5444,12 +5815,14 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, /* done with de, release bh */ brelse(bh); if (rc != 0) { - if (unlikely(ino == osd_remote_parent_ino(dev))) { + if (unlikely(is_remote_parent_ino(dev, ino))) { const char *name = (const char *)key; - /* If the parent is on remote MDT, and there + /* + * If the parent is on remote MDT, and there * is no FID-in-dirent, then we have to get - * the parent FID from the linkEA. */ + * the parent FID from the linkEA. + */ if (likely(strlen(name) == 2 && name[0] == '.' && name[1] == '.')) rc = osd_get_pfid_from_linkea(env, obj, @@ -5494,12 +5867,13 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, const struct dt_key *key, struct thandle *handle) { - struct osd_thandle *oh; - struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec; - const struct lu_fid *fid = rec1->rec_fid; - int credits, rc = 0; - struct osd_idmap_cache *idc; + struct osd_thandle *oh; + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec; + const struct lu_fid *fid = rec1->rec_fid; + int credits, rc = 0; + struct osd_idmap_cache *idc; + ENTRY; LASSERT(!dt_object_remote(dt)); @@ -5512,16 +5886,20 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, credits = osd_dto_credits_noquota[DTO_INDEX_INSERT]; - /* we can't call iget() while a transactions is running + /* + * we can't call iget() while a transactions is running * (this can lead to a deadlock), but we need to know * inum and object type. so we find this information at - * declaration and cache in per-thread info */ + * declaration and cache in per-thread info + */ idc = osd_idc_find_or_init(env, osd, fid); if (IS_ERR(idc)) RETURN(PTR_ERR(idc)); if (idc->oic_remote) { - /* a reference to remote inode is represented by an - * agent inode which we have to create */ + /* + * a reference to remote inode is represented by an + * agent inode which we have to create + */ credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE]; credits += osd_dto_credits_noquota[DTO_INDEX_INSERT]; } @@ -5531,9 +5909,11 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, if (osd_dt_obj(dt)->oo_inode != NULL) { struct inode *inode = osd_dt_obj(dt)->oo_inode; - /* We ignore block quota on meta pool (MDTs), so needn't + /* + * We ignore block quota on meta pool (MDTs), so needn't * calculate how many blocks will be consumed by this index - * insert */ + * insert + */ rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), i_projid_read(inode), 0, @@ -5543,7 +5923,8 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, RETURN(rc); #ifdef HAVE_PROJECT_QUOTA - /* Reserve credits for local agent inode to transfer + /* + * Reserve credits for local agent inode to transfer * to 0, quota enforcement is ignored in this case. */ if (idc->oic_remote && @@ -5551,11 +5932,10 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, i_projid_read(inode) != 0) rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh, 0, i_projid_read(inode), - 0, false, PRJQUOTA); + 0, false, PRJQUOTA, true); #endif } - RETURN(rc); } @@ -5575,15 +5955,16 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, const struct dt_key *key, struct thandle *th, int ignore_quota) { - struct osd_object *obj = osd_dt_obj(dt); - struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec; - const struct lu_fid *fid = rec1->rec_fid; - const char *name = (const char *)key; - struct osd_thread_info *oti = osd_oti_get(env); - struct inode *child_inode = NULL; - struct osd_idmap_cache *idc; - int rc; + struct osd_object *obj = osd_dt_obj(dt); + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec; + const struct lu_fid *fid = rec1->rec_fid; + const char *name = (const char *)key; + struct osd_thread_info *oti = osd_oti_get(env); + struct inode *child_inode = NULL; + struct osd_idmap_cache *idc; + int rc; + ENTRY; if (!dt_object_exists(dt)) @@ -5599,17 +5980,24 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, idc = osd_idc_find(env, osd, fid); if (unlikely(idc == NULL)) { - /* this dt_insert() wasn't declared properly, so - * FID is missing in OI cache. we better do not - * lookup FID in FLDB/OI and don't risk to deadlock, - * but in some special cases (lfsck testing, etc) - * it's much simpler than fixing a caller */ - CERROR("%s: "DFID" wasn't declared for insert\n", - osd_name(osd), PFID(fid)); - dump_stack(); idc = osd_idc_find_or_init(env, osd, fid); - if (IS_ERR(idc)) + if (IS_ERR(idc)) { + /* + * this dt_insert() wasn't declared properly, so + * FID is missing in OI cache. we better do not + * lookup FID in FLDB/OI and don't risk to deadlock, + * but in some special cases (lfsck testing, etc) + * it's much simpler than fixing a caller. + * + * normally this error should be placed after the first + * find, but migrate may attach source stripes to + * target, which doesn't create stripes. + */ + CERROR("%s: "DFID" wasn't declared for insert\n", + osd_name(osd), PFID(fid)); + dump_stack(); RETURN(PTR_ERR(idc)); + } } if (idc->oic_remote) { @@ -5634,6 +6022,7 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, child_inode = oti->oti_inode; if (unlikely(child_inode == NULL)) { struct ldiskfs_inode_info *lii; + OBD_ALLOC_PTR(lii); if (lii == NULL) RETURN(-ENOMEM); @@ -5666,11 +6055,11 @@ static struct dt_it *osd_it_iam_init(const struct lu_env *env, struct dt_object *dt, __u32 unused) { - struct osd_it_iam *it; - struct osd_object *obj = osd_dt_obj(dt); - struct lu_object *lo = &dt->do_lu; - struct iam_path_descr *ipd; - struct iam_container *bag = &obj->oo_dir->od_container; + struct osd_it_iam *it; + struct osd_object *obj = osd_dt_obj(dt); + struct lu_object *lo = &dt->do_lu; + struct iam_path_descr *ipd; + struct iam_container *bag = &obj->oo_dir->od_container; if (!dt_object_exists(dt)) return ERR_PTR(-ENOENT); @@ -5697,8 +6086,8 @@ static struct dt_it *osd_it_iam_init(const struct lu_env *env, */ static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di) { - struct osd_it_iam *it = (struct osd_it_iam *)di; - struct osd_object *obj = it->oi_obj; + struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_object *obj = it->oi_obj; iam_it_fini(&it->oi_it); osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd); @@ -5718,10 +6107,10 @@ static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di) */ static int osd_it_iam_get(const struct lu_env *env, - struct dt_it *di, const struct dt_key *key) + struct dt_it *di, const struct dt_key *key) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_it_iam *it = (struct osd_it_iam *)di; if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) { /* swab quota uid/gid */ @@ -5729,7 +6118,7 @@ static int osd_it_iam_get(const struct lu_env *env, key = (struct dt_key *)&oti->oti_quota_id; } - return iam_it_get(&it->oi_it, (const struct iam_key *)key); + return iam_it_get(&it->oi_it, (const struct iam_key *)key); } /** @@ -5739,9 +6128,9 @@ static int osd_it_iam_get(const struct lu_env *env, */ static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; - iam_it_put(&it->oi_it); + iam_it_put(&it->oi_it); } /** @@ -5756,9 +6145,9 @@ static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di) static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; - return iam_it_next(&it->oi_it); + return iam_it_next(&it->oi_it); } /** @@ -5766,12 +6155,12 @@ static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di) */ static struct dt_key *osd_it_iam_key(const struct lu_env *env, - const struct dt_it *di) + const struct dt_it *di) { struct osd_thread_info *oti = osd_oti_get(env); - struct osd_it_iam *it = (struct osd_it_iam *)di; - struct osd_object *obj = it->oi_obj; - struct dt_key *key; + struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_object *obj = it->oi_obj; + struct dt_key *key; key = (struct dt_key *)iam_it_key_get(&it->oi_it); @@ -5790,9 +6179,9 @@ static struct dt_key *osd_it_iam_key(const struct lu_env *env, static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; - return iam_it_key_size(&it->oi_it); + return iam_it_key_size(&it->oi_it); } static inline void @@ -5837,21 +6226,22 @@ osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset, * Return pointer to the record under iterator. */ static int osd_it_iam_rec(const struct lu_env *env, - const struct dt_it *di, - struct dt_rec *dtrec, __u32 attr) + const struct dt_it *di, + struct dt_rec *dtrec, __u32 attr) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; struct osd_thread_info *info = osd_oti_get(env); + ENTRY; if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) { const struct osd_fid_pack *rec; - struct lu_fid *fid = &info->oti_fid; - struct lu_dirent *lde = (struct lu_dirent *)dtrec; - char *name; - int namelen; - __u64 hash; - int rc; + struct lu_fid *fid = &info->oti_fid; + struct lu_dirent *lde = (struct lu_dirent *)dtrec; + char *name; + int namelen; + __u64 hash; + int rc; name = (char *)iam_it_key_get(&it->oi_it); if (IS_ERR(name)) @@ -5889,9 +6279,9 @@ static int osd_it_iam_rec(const struct lu_env *env, */ static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; - return iam_it_store(&it->oi_it); + return iam_it_store(&it->oi_it); } /** @@ -5906,31 +6296,31 @@ static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di) */ static int osd_it_iam_load(const struct lu_env *env, - const struct dt_it *di, __u64 hash) + const struct dt_it *di, __u64 hash) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; - return iam_it_load(&it->oi_it, hash); + return iam_it_load(&it->oi_it, hash); } static const struct dt_index_operations osd_index_iam_ops = { - .dio_lookup = osd_index_iam_lookup, - .dio_declare_insert = osd_index_declare_iam_insert, - .dio_insert = osd_index_iam_insert, - .dio_declare_delete = osd_index_declare_iam_delete, - .dio_delete = osd_index_iam_delete, - .dio_it = { - .init = osd_it_iam_init, - .fini = osd_it_iam_fini, - .get = osd_it_iam_get, - .put = osd_it_iam_put, - .next = osd_it_iam_next, - .key = osd_it_iam_key, - .key_size = osd_it_iam_key_size, - .rec = osd_it_iam_rec, - .store = osd_it_iam_store, - .load = osd_it_iam_load - } + .dio_lookup = osd_index_iam_lookup, + .dio_declare_insert = osd_index_declare_iam_insert, + .dio_insert = osd_index_iam_insert, + .dio_declare_delete = osd_index_declare_iam_delete, + .dio_delete = osd_index_iam_delete, + .dio_it = { + .init = osd_it_iam_init, + .fini = osd_it_iam_fini, + .get = osd_it_iam_get, + .put = osd_it_iam_put, + .next = osd_it_iam_next, + .key = osd_it_iam_key, + .key_size = osd_it_iam_key_size, + .rec = osd_it_iam_rec, + .store = osd_it_iam_store, + .load = osd_it_iam_load + } }; @@ -5944,12 +6334,13 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, struct dt_object *dt, __u32 attr) { - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thread_info *info = osd_oti_get(env); - struct osd_it_ea *oie; - struct file *file; - struct lu_object *lo = &dt->do_lu; - struct dentry *obj_dentry; + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thread_info *info = osd_oti_get(env); + struct osd_it_ea *oie; + struct file *file; + struct lu_object *lo = &dt->do_lu; + struct dentry *obj_dentry; + ENTRY; if (!dt_object_exists(dt) || obj->oo_destroyed) @@ -5975,7 +6366,7 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, if (oie->oie_buf == NULL) RETURN(ERR_PTR(-ENOMEM)); } - oie->oie_obj = obj; + oie->oie_obj = obj; file = &oie->oie_file; @@ -5990,7 +6381,7 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, set_file_inode(file, obj->oo_inode); lu_object_get(lo); - RETURN((struct dt_it *) oie); + RETURN((struct dt_it *)oie); } /** @@ -6000,10 +6391,10 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, */ static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di) { - struct osd_thread_info *info = osd_oti_get(env); - struct osd_it_ea *oie = (struct osd_it_ea *)di; - struct osd_object *obj = oie->oie_obj; - struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct osd_it_ea *oie = (struct osd_it_ea *)di; + struct osd_object *obj = oie->oie_obj; + struct inode *inode = obj->oo_inode; ENTRY; oie->oie_file.f_op->release(inode, &oie->oie_file); @@ -6027,18 +6418,18 @@ static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di) * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty(). */ static int osd_it_ea_get(const struct lu_env *env, - struct dt_it *di, const struct dt_key *key) + struct dt_it *di, const struct dt_key *key) { - struct osd_it_ea *it = (struct osd_it_ea *)di; + struct osd_it_ea *it = (struct osd_it_ea *)di; - ENTRY; - LASSERT(((const char *)key)[0] == '\0'); - it->oie_file.f_pos = 0; - it->oie_rd_dirent = 0; - it->oie_it_dirent = 0; - it->oie_dirent = NULL; + ENTRY; + LASSERT(((const char *)key)[0] == '\0'); + it->oie_file.f_pos = 0; + it->oie_rd_dirent = 0; + it->oie_it_dirent = 0; + it->oie_dirent = NULL; - RETURN(+1); + RETURN(+1); } /** @@ -6071,31 +6462,31 @@ static int osd_ldiskfs_filldir(struct dir_context *buf, static int osd_ldiskfs_filldir(void *buf, #endif const char *name, int namelen, - loff_t offset, __u64 ino, unsigned d_type) -{ - struct osd_it_ea *it = - ((struct osd_filldir_cbs *)buf)->it; - struct osd_object *obj = it->oie_obj; - struct osd_it_ea_dirent *ent = it->oie_dirent; - struct lu_fid *fid = &ent->oied_fid; - struct osd_fid_pack *rec; - ENTRY; - - /* this should never happen */ - if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) { - CERROR("ldiskfs return invalid namelen %d\n", namelen); - RETURN(-EIO); - } - - if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen > - OSD_IT_EA_BUFSIZE) - RETURN(1); + loff_t offset, __u64 ino, unsigned int d_type) +{ + struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it; + struct osd_object *obj = it->oie_obj; + struct osd_it_ea_dirent *ent = it->oie_dirent; + struct lu_fid *fid = &ent->oied_fid; + struct osd_fid_pack *rec; + + ENTRY; + +/* this should never happen */ + if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) { + CERROR("ldiskfs return invalid namelen %d\n", namelen); + RETURN(-EIO); + } + + if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen > + OSD_IT_EA_BUFSIZE) + RETURN(1); /* "." is just the object itself. */ if (namelen == 1 && name[0] == '.') { *fid = obj->oo_dt.do_lu.lo_header->loh_fid; } else if (d_type & LDISKFS_DIRENT_LUFID) { - rec = (struct osd_fid_pack*) (name + namelen + 1); + rec = (struct osd_fid_pack *)(name + namelen + 1); if (osd_fid_unpack(fid, rec) != 0) fid_zero(fid); } else { @@ -6109,16 +6500,16 @@ static int osd_ldiskfs_filldir(void *buf, *fid = obj->oo_dt.do_lu.lo_header->loh_fid; } - ent->oied_ino = ino; - ent->oied_off = offset; - ent->oied_namelen = namelen; - ent->oied_type = d_type; + ent->oied_ino = ino; + ent->oied_off = offset; + ent->oied_namelen = namelen; + ent->oied_type = d_type; - memcpy(ent->oied_name, name, namelen); + memcpy(ent->oied_name, name, namelen); - it->oie_rd_dirent++; - it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen); - RETURN(0); + it->oie_rd_dirent++; + it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen); + RETURN(0); } /** @@ -6132,14 +6523,14 @@ static int osd_ldiskfs_filldir(void *buf, * \retval +1 reach the end of entry */ static int osd_ldiskfs_it_fill(const struct lu_env *env, - const struct dt_it *di) -{ - struct osd_it_ea *it = (struct osd_it_ea *)di; - struct osd_object *obj = it->oie_obj; - struct inode *inode = obj->oo_inode; - struct htree_lock *hlock = NULL; - struct file *filp = &it->oie_file; - int rc = 0; + const struct dt_it *di) +{ + struct osd_it_ea *it = (struct osd_it_ea *)di; + struct osd_object *obj = it->oie_obj; + struct inode *inode = obj->oo_inode; + struct htree_lock *hlock = NULL; + struct file *filp = &it->oie_file; + int rc = 0; struct osd_filldir_cbs buf = { #ifdef HAVE_DIR_CONTEXT .ctx.actor = osd_ldiskfs_filldir, @@ -6147,34 +6538,40 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env, .it = it }; - ENTRY; - it->oie_dirent = it->oie_buf; - it->oie_rd_dirent = 0; + ENTRY; + it->oie_dirent = it->oie_buf; + it->oie_rd_dirent = 0; - if (obj->oo_hl_head != NULL) { - hlock = osd_oti_get(env)->oti_hlock; - ldiskfs_htree_lock(hlock, obj->oo_hl_head, - inode, LDISKFS_HLOCK_READDIR); - } else { + if (obj->oo_hl_head != NULL) { + hlock = osd_oti_get(env)->oti_hlock; + ldiskfs_htree_lock(hlock, obj->oo_hl_head, + inode, LDISKFS_HLOCK_READDIR); + } else { down_read(&obj->oo_ext_idx_sem); - } + } #ifdef HAVE_DIR_CONTEXT buf.ctx.pos = filp->f_pos; +#ifdef HAVE_ITERATE_SHARED + rc = inode->i_fop->iterate_shared(filp, &buf.ctx); +#else rc = inode->i_fop->iterate(filp, &buf.ctx); +#endif filp->f_pos = buf.ctx.pos; #else rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir); #endif - if (hlock != NULL) - ldiskfs_htree_unlock(hlock); - else + if (hlock != NULL) + ldiskfs_htree_unlock(hlock); + else up_read(&obj->oo_ext_idx_sem); if (it->oie_rd_dirent == 0) { - /*If it does not get any dirent, it means it has been reached - *to the end of the dir */ + /* + * If it does not get any dirent, it means it has been reached + * to the end of the dir + */ it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file); if (rc == 0) rc = 1; @@ -6199,26 +6596,26 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env, */ static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di) { - struct osd_it_ea *it = (struct osd_it_ea *)di; - int rc; + struct osd_it_ea *it = (struct osd_it_ea *)di; + int rc; - ENTRY; + ENTRY; - if (it->oie_it_dirent < it->oie_rd_dirent) { - it->oie_dirent = - (void *) it->oie_dirent + - cfs_size_round(sizeof(struct osd_it_ea_dirent) + - it->oie_dirent->oied_namelen); - it->oie_it_dirent++; - RETURN(0); - } else { + if (it->oie_it_dirent < it->oie_rd_dirent) { + it->oie_dirent = + (void *)it->oie_dirent + + cfs_size_round(sizeof(struct osd_it_ea_dirent) + + it->oie_dirent->oied_namelen); + it->oie_it_dirent++; + rc = 0; + } else { if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file)) - rc = +1; - else - rc = osd_ldiskfs_it_fill(env, di); - } + rc = 1; + else + rc = osd_ldiskfs_it_fill(env, di); + } - RETURN(rc); + RETURN(rc); } /** @@ -6229,15 +6626,15 @@ static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di) * \retval key i.e. struct dt_key on success */ static struct dt_key *osd_it_ea_key(const struct lu_env *env, - const struct dt_it *di) + const struct dt_it *di) { - struct osd_it_ea *it = (struct osd_it_ea *)di; + struct osd_it_ea *it = (struct osd_it_ea *)di; - return (struct dt_key *)it->oie_dirent->oied_name; + return (struct dt_key *)it->oie_dirent->oied_name; } /** - * Returns the key's size at current position from iterator's in memory structure. + * Returns key's size at current position from iterator's in memory structure. * * \param di iterator's in memory structure * @@ -6245,9 +6642,9 @@ static struct dt_key *osd_it_ea_key(const struct lu_env *env, */ static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di) { - struct osd_it_ea *it = (struct osd_it_ea *)di; + struct osd_it_ea *it = (struct osd_it_ea *)di; - return it->oie_dirent->oied_namelen; + return it->oie_dirent->oied_namelen; } static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de) @@ -6261,7 +6658,7 @@ static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de) static inline bool osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen, - unsigned blocksize, bool dotdot) + unsigned int blocksize, bool dotdot) { if (dotdot) return osd_dotdot_has_space(de); @@ -6280,17 +6677,17 @@ osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev, struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock, bool dotdot) { - struct inode *dir = dentry->d_parent->d_inode; - struct inode *inode = dentry->d_inode; - struct osd_fid_pack *rec; + struct inode *dir = dentry->d_parent->d_inode; + struct inode *inode = dentry->d_inode; + struct osd_fid_pack *rec; struct ldiskfs_dentry_param *ldp; - int namelen = dentry->d_name.len; - int rc; - struct osd_thread_info *info = osd_oti_get(env); + int namelen = dentry->d_name.len; + int rc; + struct osd_thread_info *info = osd_oti_get(env); + ENTRY; - if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb, - LDISKFS_FEATURE_INCOMPAT_DIRDATA)) + if (!ldiskfs_has_feature_dirdata(inode->i_sb)) RETURN(0); /* There is enough space to hold the FID-in-dirent. */ @@ -6320,8 +6717,10 @@ osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev, dentry->d_fsdata = (void *)ldp; ll_vfs_dq_init(dir); rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock); - /* It is too bad, we cannot reinsert the name entry back. - * That means we lose it! */ + /* + * It is too bad, we cannot reinsert the name entry back. + * That means we lose it! + */ if (rc != 0) CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, " "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n", @@ -6337,24 +6736,25 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, struct osd_it_ea *it, struct lu_fid *fid, struct osd_inode_id *id, __u32 *attr) { - struct osd_thread_info *info = osd_oti_get(env); - struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma; - struct osd_device *dev = osd_obj2dev(obj); - struct super_block *sb = osd_sb(dev); - const char *devname = osd_name(dev); - struct osd_it_ea_dirent *ent = it->oie_dirent; - struct inode *dir = obj->oo_inode; - struct htree_lock *hlock = NULL; - struct buffer_head *bh = NULL; - handle_t *jh = NULL; + struct osd_thread_info *info = osd_oti_get(env); + struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma; + struct osd_device *dev = osd_obj2dev(obj); + struct super_block *sb = osd_sb(dev); + const char *devname = osd_name(dev); + struct osd_it_ea_dirent *ent = it->oie_dirent; + struct inode *dir = obj->oo_inode; + struct htree_lock *hlock = NULL; + struct buffer_head *bh = NULL; + handle_t *jh = NULL; struct ldiskfs_dir_entry_2 *de; - struct dentry *dentry; - struct inode *inode; + struct dentry *dentry; + struct inode *inode; const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu); - int credits; - int rc; - bool dotdot = false; - bool dirty = false; + int credits; + int rc; + bool dotdot = false; + bool dirty = false; + ENTRY; if (ent->oied_name[0] == '.') { @@ -6370,8 +6770,10 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, if (IS_ERR(inode)) { rc = PTR_ERR(inode); if (rc == -ENOENT || rc == -ESTALE) { - /* Maybe dangling name entry, or - * corrupted directory entry. */ + /* + * Maybe dangling name entry, or + * corrupted directory entry. + */ *attr |= LUDA_UNKNOWN; rc = 0; } else { @@ -6394,7 +6796,8 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, else if (rc != 0) GOTO(out, rc); - /* We need to ensure that the name entry is still valid. + /* + * We need to ensure that the name entry is still valid. * Because it may be removed or renamed by other already. * * The unlink or rename operation will start journal before PDO lock, @@ -6406,7 +6809,8 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, * We may need to remove the name entry firstly, then insert back. * One credit is for user quota file update. * One credit is for group quota file update. - * Two credits are for dirty inode. */ + * Two credits are for dirty inode. + */ credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] + osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2; @@ -6428,10 +6832,12 @@ again: if (obj->oo_hl_head != NULL) { hlock = osd_oti_get(env)->oti_hlock; - /* "0" means exclusive lock for the whole directory. + /* + * "0" means exclusive lock for the whole directory. * We need to prevent others access such name entry * during the delete + insert. Neither HLOCK_ADD nor - * HLOCK_DEL cannot guarantee the atomicity. */ + * HLOCK_DEL cannot guarantee the atomicity. + */ ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0); } else { down_write(&obj->oo_ext_idx_sem); @@ -6453,12 +6859,14 @@ again: GOTO(out, rc = 0); } - /* For dotdot entry, if there is not enough space to hold the + /* + * For dotdot entry, if there is not enough space to hold the * FID-in-dirent, just keep them there. It only happens when the * device upgraded from 1.8 or restored from MDT file-level backup. * For the whole directory, only dotdot entry have no FID-in-dirent * and needs to get FID from LMA when readdir, it will not affect the - * performance much. */ + * performance much. + */ if (dotdot && !osd_dotdot_has_space(de)) { *attr |= LUDA_UNKNOWN; @@ -6475,8 +6883,10 @@ again: if (likely(dotdot && fid_seq(tfid) == FID_SEQ_LOCAL_FILE && fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) { - /* It must be REMOTE_PARENT_DIR and as the - * 'dotdot' entry of remote directory */ + /* + * It must be REMOTE_PARENT_DIR and as the + * 'dotdot' entry of remote directory + */ *attr |= LUDA_IGNORE; } else { CDEBUG(D_LFSCK, "%s: expect remote agent " @@ -6499,9 +6909,11 @@ again: if (rc == -ENOENT || (rc == -ENODATA && !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) { - /* linkEA does not recognize the dirent entry, + /* + * linkEA does not recognize the dirent entry, * it may because the dirent entry corruption - * and points to other's inode. */ + * and points to other's inode. + */ CDEBUG(D_LFSCK, "%s: the target inode does not " "recognize the dirent, dir = %lu/%u, " " name = %.*s, ino = %llu, " @@ -6526,8 +6938,10 @@ again: } if (lma != NULL) { - /* linkEA recognizes the dirent entry, the FID-in-LMA is - * valid, trusted, in spite of fid_is_sane(fid) or not. */ + /* + * linkEA recognizes the dirent entry, the FID-in-LMA is + * valid, trusted, in spite of fid_is_sane(fid) or not. + */ if (*attr & LUDA_VERIFY_DRYRUN) { *fid = lma->lma_self_fid; *attr |= LUDA_REPAIR; @@ -6591,8 +7005,10 @@ again: dirty = true; if (unlikely(fid_is_sane(fid))) { - /* FID-in-dirent exists, but FID-in-LMA is lost. - * Trust the FID-in-dirent, and add FID-in-LMA. */ + /* + * FID-in-dirent exists, but FID-in-LMA is lost. + * Trust the FID-in-dirent, and add FID-in-LMA. + */ rc = osd_ea_fid_set(info, inode, fid, 0, 0); if (rc == 0) *attr |= LUDA_REPAIR; @@ -6606,8 +7022,10 @@ again: ent->oied_ino, PFID(fid), rc); } else if (dev->od_index == 0) { lu_igif_build(fid, inode->i_ino, inode->i_generation); - /* It is probably IGIF object. Only aappend the - * FID-in-dirent. OI scrub will process FID-in-LMA. */ + /* + * It is probably IGIF object. Only aappend the + * FID-in-dirent. OI scrub will process FID-in-LMA. + */ rc = osd_dirent_reinsert(env, dev, jh, dentry, fid, bh, de, hlock, dotdot); if (rc == 0) @@ -6667,20 +7085,23 @@ static inline int osd_it_ea_rec(const struct lu_env *env, struct osd_device *dev = osd_obj2dev(obj); struct osd_thread_info *oti = osd_oti_get(env); struct osd_inode_id *id = &oti->oti_id; - struct lu_fid *fid = &it->oie_dirent->oied_fid; + struct lu_fid *fid = &it->oie_dirent->oied_fid; struct lu_dirent *lde = (struct lu_dirent *)dtrec; - __u32 ino = it->oie_dirent->oied_ino; - int rc = 0; + __u32 ino = it->oie_dirent->oied_ino; + int rc = 0; + ENTRY; - LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode); + LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino)); if (attr & LUDA_VERIFY) { - if (unlikely(ino == osd_remote_parent_ino(dev))) { + if (unlikely(is_remote_parent_ino(dev, ino))) { attr |= LUDA_IGNORE; - /* If the parent is on remote MDT, and there + /* + * If the parent is on remote MDT, and there * is no FID-in-dirent, then we have to get - * the parent FID from the linkEA. */ + * the parent FID from the linkEA. + */ if (!fid_is_sane(fid) && it->oie_dirent->oied_namelen == 2 && it->oie_dirent->oied_name[0] == '.' && @@ -6697,14 +7118,17 @@ static inline int osd_it_ea_rec(const struct lu_env *env, attr &= ~LU_DIRENT_ATTRS_MASK; if (!fid_is_sane(fid)) { bool is_dotdot = false; + if (it->oie_dirent->oied_namelen == 2 && it->oie_dirent->oied_name[0] == '.' && it->oie_dirent->oied_name[1] == '.') is_dotdot = true; - /* If the parent is on remote MDT, and there + /* + * If the parent is on remote MDT, and there * is no FID-in-dirent, then we have to get - * the parent FID from the linkEA. */ - if (ino == osd_remote_parent_ino(dev) && is_dotdot) { + * the parent FID from the linkEA. + */ + if (is_remote_parent_ino(dev, ino) && is_dotdot) { rc = osd_get_pfid_from_linkea(env, obj, fid); } else { if (is_dotdot == false && @@ -6767,9 +7191,9 @@ static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di, */ static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di) { - struct osd_it_ea *it = (struct osd_it_ea *)di; + struct osd_it_ea *it = (struct osd_it_ea *)di; - return it->oie_dirent->oied_off; + return it->oie_dirent->oied_off; } /** @@ -6783,22 +7207,22 @@ static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di) * \retval -ve on error */ static int osd_it_ea_load(const struct lu_env *env, - const struct dt_it *di, __u64 hash) + const struct dt_it *di, __u64 hash) { - struct osd_it_ea *it = (struct osd_it_ea *)di; - int rc; + struct osd_it_ea *it = (struct osd_it_ea *)di; + int rc; - ENTRY; - it->oie_file.f_pos = hash; + ENTRY; + it->oie_file.f_pos = hash; - rc = osd_ldiskfs_it_fill(env, di); + rc = osd_ldiskfs_it_fill(env, di); if (rc > 0) rc = -ENODATA; - if (rc == 0) - rc = +1; + if (rc == 0) + rc = 1; - RETURN(rc); + RETURN(rc); } /** @@ -6812,18 +7236,18 @@ static int osd_it_ea_load(const struct lu_env *env, static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt, struct dt_rec *rec, const struct dt_key *key) { - struct osd_object *obj = osd_dt_obj(dt); - int rc = 0; + struct osd_object *obj = osd_dt_obj(dt); + int rc = 0; - ENTRY; + ENTRY; - LASSERT(S_ISDIR(obj->oo_inode->i_mode)); - LINVRNT(osd_invariant(obj)); + LASSERT(S_ISDIR(obj->oo_inode->i_mode)); + LINVRNT(osd_invariant(obj)); - rc = osd_ea_lookup_rec(env, obj, rec, key); - if (rc == 0) - rc = +1; - RETURN(rc); + rc = osd_ea_lookup_rec(env, obj, rec, key); + if (rc == 0) + rc = 1; + RETURN(rc); } /** @@ -6852,39 +7276,39 @@ static const struct dt_index_operations osd_index_ea_ops = { }; static void *osd_key_init(const struct lu_context *ctx, - struct lu_context_key *key) + struct lu_context_key *key) { - struct osd_thread_info *info; + struct osd_thread_info *info; - OBD_ALLOC_PTR(info); - if (info == NULL) - return ERR_PTR(-ENOMEM); + OBD_ALLOC_PTR(info); + if (info == NULL) + return ERR_PTR(-ENOMEM); - OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); - if (info->oti_it_ea_buf == NULL) - goto out_free_info; + OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); + if (info->oti_it_ea_buf == NULL) + goto out_free_info; - info->oti_env = container_of(ctx, struct lu_env, le_ctx); + info->oti_env = container_of(ctx, struct lu_env, le_ctx); - info->oti_hlock = ldiskfs_htree_lock_alloc(); - if (info->oti_hlock == NULL) - goto out_free_ea; + info->oti_hlock = ldiskfs_htree_lock_alloc(); + if (info->oti_hlock == NULL) + goto out_free_ea; - return info; + return info; - out_free_ea: - OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); - out_free_info: - OBD_FREE_PTR(info); - return ERR_PTR(-ENOMEM); +out_free_ea: + OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); +out_free_info: + OBD_FREE_PTR(info); + return ERR_PTR(-ENOMEM); } static void osd_key_fini(const struct lu_context *ctx, - struct lu_context_key *key, void* data) + struct lu_context_key *key, void *data) { struct osd_thread_info *info = data; struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode); - struct osd_idmap_cache *idc = info->oti_ins_cache; + struct osd_idmap_cache *idc = info->oti_ins_cache; if (info->oti_inode != NULL) OBD_FREE_PTR(lli); @@ -6904,41 +7328,42 @@ static void osd_key_fini(const struct lu_context *ctx, } static void osd_key_exit(const struct lu_context *ctx, - struct lu_context_key *key, void *data) + struct lu_context_key *key, void *data) { - struct osd_thread_info *info = data; + struct osd_thread_info *info = data; - LASSERT(info->oti_r_locks == 0); - LASSERT(info->oti_w_locks == 0); - LASSERT(info->oti_txns == 0); + LASSERT(info->oti_r_locks == 0); + LASSERT(info->oti_w_locks == 0); + LASSERT(info->oti_txns == 0); } /* type constructor/destructor: osd_type_init, osd_type_fini */ LU_TYPE_INIT_FINI(osd, &osd_key); struct lu_context_key osd_key = { - .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL, - .lct_init = osd_key_init, - .lct_fini = osd_key_fini, - .lct_exit = osd_key_exit + .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL, + .lct_init = osd_key_init, + .lct_fini = osd_key_fini, + .lct_exit = osd_key_exit }; static int osd_device_init(const struct lu_env *env, struct lu_device *d, - const char *name, struct lu_device *next) + const char *name, struct lu_device *next) { struct osd_device *osd = osd_dev(d); - if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) - >= sizeof(osd->od_svname)) + if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >= + sizeof(osd->od_svname)) return -E2BIG; return osd_procfs_init(osd, name); } static int osd_fid_init(const struct lu_env *env, struct osd_device *osd) { - struct seq_server_site *ss = osd_seq_site(osd); - int rc; + struct seq_server_site *ss = osd_seq_site(osd); + int rc; + ENTRY; if (osd->od_is_ost || osd->od_cl_seq != NULL) @@ -6960,9 +7385,11 @@ static int osd_fid_init(const struct lu_env *env, struct osd_device *osd) } if (ss->ss_node_id == 0) { - /* If the OSD on the sequence controller(MDT0), then allocate + /* + * If the OSD on the sequence controller(MDT0), then allocate * sequence here, otherwise allocate sequence after connected - * to MDT0 (see mdt_register_lwp_callback()). */ + * to MDT0 (see mdt_register_lwp_callback()). + */ rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv, &osd->od_cl_seq->lcs_space, env); } @@ -6986,8 +7413,10 @@ static int osd_shutdown(const struct lu_env *env, struct osd_device *o) /* shutdown quota slave instance associated with the device */ if (o->od_quota_slave != NULL) { - qsd_fini(env, o->od_quota_slave); + struct qsd_instance *qsd = o->od_quota_slave; + o->od_quota_slave = NULL; + qsd_fini(env, qsd); } osd_fid_fini(env, o); @@ -7012,21 +7441,22 @@ static void osd_umount(const struct lu_env *env, struct osd_device *o) } static int osd_mount(const struct lu_env *env, - struct osd_device *o, struct lustre_cfg *cfg) + struct osd_device *o, struct lustre_cfg *cfg) { - const char *name = lustre_cfg_string(cfg, 0); - const char *dev = lustre_cfg_string(cfg, 1); - const char *opts; - unsigned long page, s_flags, lmd_flags = 0; - struct page *__page; + const char *name = lustre_cfg_string(cfg, 0); + const char *dev = lustre_cfg_string(cfg, 1); + const char *opts; + unsigned long page, s_flags, lmd_flags = 0; + struct page *__page; struct file_system_type *type; - char *options = NULL; - char *str; - struct osd_thread_info *info = osd_oti_get(env); - struct lu_fid *fid = &info->oti_fid; - struct inode *inode; - int rc = 0, force_over_512tb = 0; - ENTRY; + char *options = NULL; + char *str; + struct osd_thread_info *info = osd_oti_get(env); + struct lu_fid *fid = &info->oti_fid; + struct inode *inode; + int rc = 0, force_over_512tb = 0; + + ENTRY; if (o->od_mnt != NULL) RETURN(0); @@ -7094,6 +7524,7 @@ static int osd_mount(const struct lu_env *env, strcat(options, opts); for (rc = 0, str = options; sout[rc]; ) { char *op = strstr(str, sout[rc]); + if (op == NULL) { rc++; str = options; @@ -7164,15 +7595,13 @@ static int osd_mount(const struct lu_env *env, GOTO(out_mnt, rc = -EROFS); } - if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb, - LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) { + if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) { CERROR("%s: device %s is mounted w/o journal\n", name, dev); GOTO(out_mnt, rc = -EINVAL); } #ifdef LDISKFS_MOUNT_DIRDATA - if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb, - LDISKFS_FEATURE_INCOMPAT_DIRDATA)) + if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb)) LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA; else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */ CWARN("%s: device %s was upgraded from Lustre-1.x without " @@ -7208,8 +7637,10 @@ static struct lu_device *osd_device_fini(const struct lu_env *env, struct lu_device *d) { struct osd_device *o = osd_dev(d); + ENTRY; + osd_index_backup(env, o, false); osd_shutdown(env, o); osd_procfs_fini(o); osd_obj_map_fini(o); @@ -7222,13 +7653,13 @@ static int osd_device_init0(const struct lu_env *env, struct osd_device *o, struct lustre_cfg *cfg) { - struct lu_device *l = osd2lu_dev(o); + struct lu_device *l = osd2lu_dev(o); struct osd_thread_info *info; - int rc; - int cplen = 0; + int rc; + int cplen = 0; /* if the module was re-loaded, env can loose its keys */ - rc = lu_env_refill((struct lu_env *) env); + rc = lu_env_refill((struct lu_env *)env); if (rc) GOTO(out, rc); info = osd_oti_get(env); @@ -7240,6 +7671,10 @@ static int osd_device_init0(const struct lu_env *env, spin_lock_init(&o->od_osfs_lock); mutex_init(&o->od_otable_mutex); INIT_LIST_HEAD(&o->od_orphan_list); + INIT_LIST_HEAD(&o->od_index_backup_list); + INIT_LIST_HEAD(&o->od_index_restore_list); + spin_lock_init(&o->od_lock); + o->od_index_backup_policy = LIBP_NONE; o->od_read_cache = 1; o->od_writethrough_cache = 1; @@ -7253,6 +7688,7 @@ static int osd_device_init0(const struct lu_env *env, GOTO(out, rc); } + o->od_index_backup_stop = 0; o->od_index = -1; /* -1 means index is invalid */ rc = server_name2index(o->od_svname, &o->od_index, NULL); if (rc == LDD_F_SV_TYPE_OST) @@ -7279,7 +7715,9 @@ static int osd_device_init0(const struct lu_env *env, INIT_LIST_HEAD(&o->od_ios_list); /* setup scrub, including OI files initialization */ + o->od_in_init = 1; rc = osd_scrub_setup(env, o); + o->od_in_init = 0; if (rc < 0) GOTO(out_site, rc); @@ -7319,11 +7757,11 @@ out: } static struct lu_device *osd_device_alloc(const struct lu_env *env, - struct lu_device_type *t, - struct lustre_cfg *cfg) + struct lu_device_type *t, + struct lustre_cfg *cfg) { struct osd_device *o; - int rc; + int rc; OBD_ALLOC_PTR(o); if (o == NULL) @@ -7331,8 +7769,10 @@ static struct lu_device *osd_device_alloc(const struct lu_env *env, rc = dt_device_init(&o->od_dt_dev, t); if (rc == 0) { - /* Because the ctx might be revived in dt_device_init, - * refill the env here */ + /* + * Because the ctx might be revived in dt_device_init, + * refill the env here + */ lu_env_refill((struct lu_env *)env); rc = osd_device_init0(env, o, cfg); if (rc) @@ -7346,10 +7786,11 @@ static struct lu_device *osd_device_alloc(const struct lu_env *env, } static struct lu_device *osd_device_free(const struct lu_env *env, - struct lu_device *d) + struct lu_device *d) { - struct osd_device *o = osd_dev(d); - ENTRY; + struct osd_device *o = osd_dev(d); + + ENTRY; /* XXX: make osd top device in order to release reference */ d->ld_site->ls_top_dev = d; @@ -7359,16 +7800,17 @@ static struct lu_device *osd_device_free(const struct lu_env *env, lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer); } lu_site_fini(&o->od_site); - dt_device_fini(&o->od_dt_dev); - OBD_FREE_PTR(o); - RETURN(NULL); + dt_device_fini(&o->od_dt_dev); + OBD_FREE_PTR(o); + RETURN(NULL); } static int osd_process_config(const struct lu_env *env, - struct lu_device *d, struct lustre_cfg *cfg) + struct lu_device *d, struct lustre_cfg *cfg) { - struct osd_device *o = osd_dev(d); - int rc; + struct osd_device *o = osd_dev(d); + int rc; + ENTRY; switch (cfg->lcfg_command) { @@ -7376,6 +7818,11 @@ static int osd_process_config(const struct lu_env *env, rc = osd_mount(env, o, cfg); break; case LCFG_CLEANUP: + /* + * For the case LCFG_PRE_CLEANUP is not called in advance, + * that may happend if hit failure during mount process. + */ + osd_index_backup(env, o, false); lu_dev_del_linkage(d->ld_site, d); rc = osd_shutdown(env, o); break; @@ -7391,6 +7838,12 @@ static int osd_process_config(const struct lu_env *env, rc = 0; } break; + case LCFG_PRE_CLEANUP: + osd_scrub_stop(o); + osd_index_backup(env, o, + o->od_index_backup_policy != LIBP_NONE); + rc = 0; + break; default: rc = -ENOSYS; } @@ -7399,17 +7852,20 @@ static int osd_process_config(const struct lu_env *env, } static int osd_recovery_complete(const struct lu_env *env, - struct lu_device *d) + struct lu_device *d) { - struct osd_device *osd = osd_dev(d); - int rc = 0; + struct osd_device *osd = osd_dev(d); + int rc = 0; + ENTRY; if (osd->od_quota_slave == NULL) RETURN(0); - /* start qsd instance on recovery completion, this notifies the quota - * slave code that we are about to process new requests now */ + /* + * start qsd instance on recovery completion, this notifies the quota + * slave code that we are about to process new requests now + */ rc = qsd_start(env, osd->od_quota_slave); RETURN(rc); } @@ -7421,9 +7877,10 @@ static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp, struct obd_device *obd, struct obd_uuid *cluuid, struct obd_connect_data *data, void *localdata) { - struct osd_device *osd = osd_dev(obd->obd_lu_dev); - struct lustre_handle conn; - int rc; + struct osd_device *osd = osd_dev(obd->obd_lu_dev); + struct lustre_handle conn; + int rc; + ENTRY; CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects); @@ -7449,7 +7906,8 @@ static int osd_obd_disconnect(struct obd_export *exp) { struct obd_device *obd = exp->exp_obd; struct osd_device *osd = osd_dev(obd->obd_lu_dev); - int rc, release = 0; + int rc, release = 0; + ENTRY; /* Only disconnect the underlying layers on the final disconnect. */ @@ -7469,10 +7927,11 @@ static int osd_obd_disconnect(struct obd_export *exp) static int osd_prepare(const struct lu_env *env, struct lu_device *pdev, struct lu_device *dev) { - struct osd_device *osd = osd_dev(dev); - struct lr_server_data *lsd = + struct osd_device *osd = osd_dev(dev); + struct lr_server_data *lsd = &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd; - int result = 0; + int result = 0; + ENTRY; if (osd->od_quota_slave != NULL) { @@ -7511,40 +7970,40 @@ static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp, } static const struct lu_object_operations osd_lu_obj_ops = { - .loo_object_init = osd_object_init, - .loo_object_delete = osd_object_delete, - .loo_object_release = osd_object_release, - .loo_object_free = osd_object_free, - .loo_object_print = osd_object_print, - .loo_object_invariant = osd_object_invariant + .loo_object_init = osd_object_init, + .loo_object_delete = osd_object_delete, + .loo_object_release = osd_object_release, + .loo_object_free = osd_object_free, + .loo_object_print = osd_object_print, + .loo_object_invariant = osd_object_invariant }; const struct lu_device_operations osd_lu_ops = { - .ldo_object_alloc = osd_object_alloc, - .ldo_process_config = osd_process_config, - .ldo_recovery_complete = osd_recovery_complete, - .ldo_prepare = osd_prepare, + .ldo_object_alloc = osd_object_alloc, + .ldo_process_config = osd_process_config, + .ldo_recovery_complete = osd_recovery_complete, + .ldo_prepare = osd_prepare, }; static const struct lu_device_type_operations osd_device_type_ops = { - .ldto_init = osd_type_init, - .ldto_fini = osd_type_fini, + .ldto_init = osd_type_init, + .ldto_fini = osd_type_fini, - .ldto_start = osd_type_start, - .ldto_stop = osd_type_stop, + .ldto_start = osd_type_start, + .ldto_stop = osd_type_stop, - .ldto_device_alloc = osd_device_alloc, - .ldto_device_free = osd_device_free, + .ldto_device_alloc = osd_device_alloc, + .ldto_device_free = osd_device_free, - .ldto_device_init = osd_device_init, - .ldto_device_fini = osd_device_fini + .ldto_device_init = osd_device_init, + .ldto_device_fini = osd_device_fini }; static struct lu_device_type osd_device_type = { - .ldt_tags = LU_DEVICE_DT, + .ldt_tags = LU_DEVICE_DT, .ldt_name = LUSTRE_OSD_LDISKFS_NAME, - .ldt_ops = &osd_device_type_ops, - .ldt_ctx_tags = LCT_LOCAL, + .ldt_ops = &osd_device_type_ops, + .ldt_ctx_tags = LCT_LOCAL, }; static int osd_health_check(const struct lu_env *env, struct obd_device *obd) @@ -7584,9 +8043,8 @@ static int __init osd_init(void) #ifdef CONFIG_KALLSYMS priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly"); - priv_dev_check_rdonly = (void *)kallsyms_lookup_name("dev_check_rdonly"); - /* Clear readonly is unused at this time */ - /*priv_dev_clear_rdonly = (void *)kallsyms_lookup_name("dev_clear_rdonly");*/ + priv_dev_check_rdonly = + (void *)kallsyms_lookup_name("dev_check_rdonly"); #endif rc = class_register_type(&osd_obd_device_ops, NULL, true,